All tags
Topic: "long-context"
not much happened today
nemotron-3-ultra nemotron-3.5-asr claude-opus-4 mythos-preview nvidia anthropic togethercompute baseten modal vllm_project fireworksai_hq ollama wandb cline primeintellect nousresearch mixture-of-experts long-context model-quantization agentic-ai streaming-speech asr low-precision-training benchmarking recursive-self-improvement code-generation model-speedup piotrz_zelasko
NVIDIA released Nemotron 3 Ultra, a fully open 550B MoE model with 55B active parameters and 1M context, optimized for long-running agent tasks with up to 5x speedup and 30% cost reduction. It features hybrid Mamba/attention, LatentMoE, native MTP, and was pretrained on 20T tokens using NVFP4 low-precision format. Benchmarks show strong performance with 47.7 Intelligence Index and 400+ output tokens/sec. The model is supported across major serving platforms. Additionally, Nemotron 3.5 ASR is an open streaming ASR model with 0.6B parameters, supporting 40 language-locale combinations and sub-100ms latency, designed for voice agents.
Anthropic highlighted early signs of recursive self-improvement (RSI) in AI, with Claude models authoring 80%+ of merged code and engineers shipping 8x more code. Claude Opus 4 achieved 3x speedup on training scripts, while Mythos Preview reached ~52x speedup and provided better research suggestions than humans 64% of the time.
not much happened today
claude-opus-4.8 gpt-5.5 qwen kimi deepseek anthropic huggingface langchain vllm_project reinforcement-learning tokenization agentic-ai api model-optimization long-context rust performance-optimization multi-agent-systems prompt-engineering jeremyphoward leo_linsky clementdelangue johnschulman2 omarsar0 hwchase17 ofirpress scaling01
Anthropic rolled out Claude Opus 4.8, which shows incremental improvements but mixed benchmark results, including better cooperation and coding behavior but some regressions in document parsing. Platform updates include mid-conversation system instructions enhancing long agent sessions, though API pricing remains a concern. A Hugging Face analysis revealed a critical bug in multi-turn reinforcement learning training loops involving tokenization mismatches, with a proposed "Token-In, Token-Out" fix. Agent harness design is evolving as a key optimization area, with LangChain's Deep Agents v0.6 achieving strong performance at much lower cost, and vllm_project releasing native weight syncing APIs and a Rust BPE tokenizer to improve tokenization efficiency. Debate continues on the value of multi-agent systems, with some seeing them as speedups and others expecting capability breakthroughs.
Anthropic raises $65B in Series H at a $965B post-money valuation, releases Opus 4.8 and Dynamic Workflows
claude-opus-4.8 claude-opus-4.7 gpt-5.5 anthropic altimeter dragoneer greenoaks sequoia andonlabs model-release reinforcement-learning agentic-ai model-evaluation long-context model-optimization fine-tuning multitasking parallel-processing dan_shipper scaling01 zephyr_z9 teortaxes_tex kimmonismus
Anthropic announced a massive $65B Series H financing at a $965B valuation, led by Altimeter, Dragoneer, Greenoaks, and Sequoia, with run-rate revenue surpassing $47B. They launched Claude Opus 4.8, an update to Opus 4.7 featuring "sharper judgment," "more honesty," and longer autonomous work at the same price. Anthropic also introduced Dynamic Workflows in Claude Code, enabling orchestration of hundreds of parallel subagents for large tasks, available in research preview across multiple platforms. Opinions on Opus 4.8 vary, with some praising it as a major leap and others viewing it as incremental or catch-up to OpenAI's GPT-5.5 family.
not much happened today
eagle-3.1 unigram-tokenizer qwen-3.5 deepseek-v4-pro mimo deep-agents-v0.6 397b-parameter-model eaglecorp vllm_project perplexity_ai alibaba lightseek nvidia mooncake flashattention kimmonismus deepseek xiaomi langchain baseten trajectory clay harvey decagon mercor rogo rlm inference-optimization long-context speculative-decoding tokenization attention-mechanisms kv-cache cache-hierarchy agent-engineering model-harness-memory-fit continual-learning quantization autoscaling memory-centric-agents evaluation-automation kimmonismus _luofuli vtrivedy10
Inference optimization is increasingly architectural, with EAGLE 3.1 improving speculative decoding and long-context handling, collaborating with vLLM and TorchSpec. Perplexity open-sourced a rebuilt Unigram tokenizer cutting CPU use by 5โ6ร and achieving 63 ยตs at 514 tokens. Qwen3.5 hits 580 tokens/s via joint efforts from Alibaba, LightSeek, NVIDIA, Mooncake, and FlashAttention-4 contributors. Price cuts in APIs from Chinese labs are sustainable due to structural KV-cache and attention improvements, exemplified by DeepSeek V4-Pro and Xiaomi MiMo reducing caching costs significantly.
Agent engineering shifts focus from model quality to model-harness-memory fit, with LangChain releasing Deep Agents v0.6 and tools like LangSmith Engine automating evaluation loops. Trajectory launched a continual learning platform with $15M funding and partners like Clay and Harvey, supporting large models including a 397B-parameter model deployed on autoscaled H100 infrastructure. Open-source memory-centric agents and minimal training harnesses also gained attention.
not much happened today
raev2 gated-deltanet-2 kda mamba-3 dclm nvidia openai nous-research representation-learning tokenization linear-attention long-context mechanistic-interpretability math data-filtering agent-infrastructure language-modeling commonsense-reasoning 1jaskiratsingh recatm sainingxie ahatamiz1 rasbt nousresearch tatsu_hashimoto goodfireai markchen90 wtgowers memecrashes cloneofsimo lvwerra
RAEv2 advances representation-first tokenization with >10x faster convergence and improved generation, tested on text-to-image and world models. NVIDIA's Gated DeltaNet-2 innovates linear attention with channel-wise gates, outperforming KDA and Mamba-3 at 1.3B parameters on language modeling and reasoning tasks. Studies on subword tokenization reveal only some benefits at scale, while data filtering research suggests that with enough compute, no filtering may be optimal at around 1e30 FLOPs. Mechanistic interpretability updates propose clustering features by joint firing patterns for better geometry understanding. OpenAI's AI-assisted breakthrough on an Erdลs unit-distance math problem sparks debate on AI's role in mathematical research. Harnesses remain key for capability improvements in agent infrastructure.
not much happened today
command-a+ claude-3.7-sonnet openai cohere reinforcement-learning reasoning multimodality model-architecture model-optimization model-releases benchmarking long-context model-efficiency transformers wtgowers hongxunwu aidangomez nickfrosst clementdelangue eliebakouch rasbt sama
OpenAI achieved a major math breakthrough by disproving a long-standing Erdลs unit distance problem using a general-purpose reasoning model, marking a milestone in AI-driven formal science and long-horizon reasoning. The result was validated by prominent mathematicians like Timothy Gowers and OpenAI researcher Hongxun Wu, highlighting the model's advanced reasoning capabilities beyond prior AI math achievements. Meanwhile, Cohere released Command A+ as an open-source Apache 2.0 licensed model, featuring a 218B MoE / 25B active multimodal architecture supporting 48 languages and optimized for low hardware requirements, runnable on as little as 2ร H100 GPUs. Benchmarks place Command A+ near Claude 4.5 Haiku in intelligence with strong non-hallucination but weaker scientific reasoning and coding. The architecture includes novel elements like a parallel transformer block, shared experts, and LayerNorm over RMSNorm.
DeepSeek v4
deepseek-v4 deepseek-v4-pro deepseek-v4-flash kimi-k2.6 glm-5.1 xiaomi-mimo-v2.5-pro gpt-5.5 gpt-5.5-pro deepseek nvidia openai lambdaapi togethercompute xiaomi long-context mixture-of-experts model-quantization memory-optimization hardware-model-co-design inference-speed agent-integration token-efficiency model-deployment open-weights reasoning hallucination-detection scaling01 ben_burtenshaw artificialanlys
DeepSeek-V4 technical release features a 1.6T-parameter MoE with 49B active parameters and 1M-token context, showcasing hybrid attention and compressed KV schemes for major memory reductions. It ranks as the #2 open-weights reasoning model behind Kimi K2.6 but has a high hallucination rate and higher serving costs. Hardware-model co-design is emphasized, with NVIDIA Blackwell Ultra delivering 150+ TPS/user and support for FP4 and FP8 quantization enabling deployment on single nodes. Positioning among open Chinese models is competitive with GLM-5.1 and Xiaomi MiMo V2.5 Pro. Meanwhile, OpenAI launched GPT-5.5 and GPT-5.5 Pro APIs with a 1M context window, focusing on improved long-running workflows and token efficiency, quickly integrated into tools like GitHub Copilot and Cursor. "GPT-5.5 handles complex, tool-heavy, ambiguous workflows with fewer retries," highlighting rapid distribution and agent integration.
not much happened today
qwen3.6-27b qwen3.5-397b-a17b privacy-filter mimo-v2.5-pro mimo-v2.5 gemini-3.1-pro gemini-3.1-flash-image alibaba openai xiaomi google google-deepmind vllm_project unsloth ggml ollama arena nous-research open-models multimodality vision tokenization pii-detection privacy enterprise-ai agentic-ai benchmarking long-context model-deployment hardware-optimization model-integration software-engineering alibaba_qwen clementdelangue altryne eliebakouch mervenoyann xiaomimo sundarpichai scaling01
Alibaba released Qwen3.6-27B, a dense, Apache 2.0 open coding model with thinking and non-thinking modes, outperforming the larger Qwen3.5-397B-A17B on multiple coding benchmarks including SWE-bench and Terminal-Bench. It supports native vision-language reasoning over images and video, with immediate ecosystem support from vLLM, Unsloth, ggml, and Ollama. OpenAI open-sourced a practical Privacy Filter model for PII detection and masking, a 1.5B parameter token-classification model with a 128k context window aimed at enterprise redaction tasks. Xiaomi announced MiMo-V2.5-Pro and MiMo-V2.5 models, emphasizing software engineering advances, long-horizon agents, and large context windows (up to 1M tokens), with strong benchmark results and integrations with Hermes and Nous. At Google Cloud Next, Google and Google DeepMind unveiled 8th-gen TPUs (TPU 8t for training and TPU 8i for inference) with claims of scaling to a million TPUs in a cluster, and launched the Gemini Enterprise Agent Platform evolving Vertex AI with Agent Studio and access to 200+ models including Gemini 3.1 Pro and Gemini 3.1 Flash Image. This marks a significant vertical integration of hardware, models, and enterprise tooling.
not much happened today
kimi-k2.6 qwen-3.6-max-preview moonshot alibaba vllm openrouter cloudflare baseten mlx nous-research opencode ollama mixture-of-experts multimodality int4-quantization long-context agentic-coding multi-agent-systems model-orchestration memory-consolidation llm-driven-replanning dynamic-context-injection
Moonshot's Kimi K2.6 is a major open-weight 1T-parameter MoE model featuring 32B active parameters, 384 experts, MLA attention, 256K context window, native multimodality, and INT4 quantization. It supports day-0 integration with platforms like vLLM, OpenRouter, Cloudflare Workers AI, and others, showcasing state-of-the-art performance on benchmarks such as HLE w/ tools 54.0, SWE-Bench Pro 58.6, and Math Vision w/ python 93.2. The model excels in long-horizon execution with over 4,000 tool calls, 12+ hour continuous runs, and 300 parallel sub-agents. Meanwhile, Alibaba's Qwen3.6-Max-Preview previewed enhanced agentic coding, improved world knowledge, and instruction following, with notable performance on AIME 2026 #15 and ranking in Code Arena. Hermes Agent is rapidly expanding its ecosystem, surpassing 100K GitHub stars and integrating with tools like Ollama and Copilot CLI, while pioneering advanced multi-agent orchestration techniques such as stateless ephemeral units, LLM-driven replanning, and dynamic context injection. These developments highlight the competitive momentum of Chinese open and semi-open labs in coding and agent models.
Anthropic's Claude Opus 4.7
claude-opus-4.7 codex gpt-rosalind anthropic openai cursor replit perplexity-ai microsoft coding agentic-ai tokenization long-context benchmarking image-processing software-engineering computer-use plugin-integration multi-terminal-support ssh-access model-expansion bcherny kimmonismus scaling01 valsai artificialanlys natolambert nrehiew_
Anthropic launched Claude Opus 4.7, its most capable Opus model yet, featuring stronger coding and agentic performance, a new tokenizer, and improved long-context handling with a new xhigh reasoning tier. Benchmarks show substantial gains, including SWE-bench Pro 64.3%, SWE-bench Verified 87.6%, and TerminalBench 69.4%, with top rankings on Vals Index and GDPval-AA. Technical changes include a new tokenizer and increased image input resolution to 3.75MP. Some long-context benchmarks showed mixed results, with a shift in focus from MRCR to Graphwalks. Adoption was rapid across tools like Cursor, VS Code, Replit Agent, and Perplexity. Meanwhile, OpenAI expanded Codex into a broader computer agent with Mac computer use, in-app browser, image generation/editing, 90+ plugins, multi-terminal support, SSH remote devbox access, and richer file previews. A new vertical life-sciences model, GPT-Rosalind, was also introduced.
Gemma 4
gemma-4 gemma-4-31b gemma-4-26b-a4b google-deepmind multimodality long-context model-architecture moe local-inference model-optimization function-calling quantization jeffdean _philschmid rasbt ggerganov clattner_llvm julien_c clementdelangue
Google DeepMind released Gemma 4, a family of open-weight, multimodal models with long-context support up to 256K tokens under an Apache 2.0 license, marking a major capability and licensing shift. The lineup includes 31B dense, 26B MoE (A4B), and two edge models (E4B, E2B) optimized for local and edge deployment with native multimodal support (text, vision, audio). Early benchmarks show Gemma-4-31B ranking #3 among open models and strong scientific reasoning performance with 85.7% GPQA Diamond. Day-0 ecosystem support includes llama.cpp, Ollama, vLLM, and LM Studio, with notable local inference performance on hardware like M2 Ultra and RTX 4090. The architecture features hybrid attention and MoE layering, diverging from standard transformers. Community and developer engagement is high, with rapid adoption and tooling integration.
not much happened today
opus-4.6 glm-5 anthropic ibm perplexity-ai llamaindex deepseek google-chrome persistent-memory agent-infrastructure cross-device-synchronization long-context sparse-attention inference-optimization computer-architecture task-completion systems-performance pamelafox tadasayy llama_index bromann dair_ai omarsar0 abxxai teknuim bcherny kimmonismus _catwu alexalbert__ realyushibai
MCP tools remain relevant for deterministic APIs despite ergonomic criticisms, with new web MCP support in Chrome v146 enabling continuous browsing agents. Persistent memory is emerging as a key differentiator for agents, with IBM improving task completion rates and multi-agent memory framed as a computer architecture challenge. Agent UX is evolving towards always-on, cross-device operation, exemplified by Perplexity Computer on iOS and Claude Code session management. Anthropic released Opus 4.6 1M context as default with no extra long-context API charges, achieving 78.3% on MRCR v2 at 1M tokens. Sparse attention optimizations like IndexCache in DeepSeek Sparse Attention yield significant speedups on large models with minimal code changes.
GPT 5.4: SOTA Knowledge Work -and- Coding -and- CUA Model, OpenAI is so very back
gpt-5.4 gpt-5.4-pro openai cursor_ai perplexity_ai arena native-computer-use long-context efficiency steering benchmarking gpu-kernels attention-mechanisms algorithmic-optimization pipeline-optimization sama reach_vb scaling01 danshipper yuchenj_uw
OpenAI launched GPT-5.4 and GPT-5.4 Pro with unified mainline and Codex models, featuring native computer use, up to ~1M token context, and efficiency improvements including a new Codex
/fast mode. Benchmarks showed strong results like OSWorld-Verified 75.0% surpassing human baseline and GDPval 83% against industry pros. User feedback highlighted coding utility but raised concerns about pricing and overthinking. Integration with devtools like Cursor, Perplexity, and Arena was announced. In systems research, FlashAttention-4 (FA4) was introduced with near-matmul speed attention on Blackwell GPUs, featuring innovations like polynomial exp emulation and online softmax. "Steering mid-response" and "fewer tokens, faster speed" were emphasized as UX and efficiency improvements. not much happened today
qwen-3.5-0.8b qwen-3.5-2b qwen-3.5-4b qwen-3.5-9b codex-5.3 claude-3 alibaba ollama lm-studio openai anthropic multimodality reinforcement-learning long-context hybrid-attention on-device-ai model-deployment agent-reliability agent-observability coding-agents benchmarking runtime-optimization token-efficiency nrehiew_ kimmonismus lioronai danielhanchen theo htihle teortaxestex theprimeagen yuchenj_uw _lewtun saen_dev _philschmid omarsar0
Alibaba released the Qwen 3.5 series with models ranging from 0.8B to 9B parameters, featuring native multimodality, scaled reinforcement learning, and targeting edge and lightweight agent deployments. The models support very long context windows up to 262K tokens (extendable to 1M) and use a novel Gated DeltaNet hybrid attention architecture combining linear and full attention layers. Deployment examples include Ollama and LM Studio, with a notable 6-bit on-device demo on iPhone 17 Pro. Evaluators are cautioned that reasoning is disabled by default on smaller models. In coding agents, Codex 5.3 shows promising benchmark results on WeirdML with 79.3% accuracy, though availability and downtime remain critical challenges, especially highlighted by Claude outages. Agent reliability and observability are emphasized as cross-functional problems requiring clear success criteria and practical evaluation strategies. Studies show that using AGENTS.md and SKILL.md guardrails can significantly reduce runtime and token usage by mitigating worst-case thrashing in coding workflows.
Claude Sonnet 4.6: clean upgrade of 4.5, mostly better with some caveats
claude-3-sonnet-4.6 claude-3-sonnet-4.5 claude-3-opus-4.5 claude-3-opus-4.6 anthropic cursor microsoft perplexity-ai cognition long-context agent-planning knowledge-work benchmarking tokenization model-integration code-execution model-updates aesthetic-quality alexalbert__ scaling01 rishdotblog claudeai kimmonismus artificialanlys
Anthropic launched Claude Sonnet 4.6, an upgrade over Sonnet 4.5, featuring broad improvements in coding, long-context reasoning, agent planning, knowledge work, and design, plus a 1M-token context window (beta). Benchmarks show Sonnet 4.6 leading on GDPval-AA ELO 1633, with significant token usage increases and improved output aesthetics. Integrations include Cursor, Windsurf, Microsoft Foundry, and Perplexity Pro/Max. Early user feedback noted some regression issues that were later fixed. Pricing remains the same as Sonnet 4.5. Tooling enhancements include code execution for filtering results, improving accuracy and efficiency.
Qwen3.5-397B-A17B: the smallest Open-Opus class, very efficient model
qwen3.5-397b-a17b qwen3.5-plus qwen3-max qwen3-vl kimi alibaba openai deepseek z-ai minimax kimi unsloth ollama vllm native-multimodality spatial-intelligence sparse-moe long-context model-quantization model-architecture model-deployment inference-optimization apache-2.0-license pete_steinberger justinlin610
Alibaba released Qwen3.5-397B-A17B, an open-weight model featuring native multimodality, spatial intelligence, and a hybrid linear attention + sparse MoE architecture supporting 201 languages and long context windows up to 256K tokens. The model shows improvements over previous versions like Qwen3-Max and Qwen3-VL, with a sparsity ratio of about 4.3%. Community discussions highlighted the Gated Delta Networks enabling efficient inference despite large model size (~800GB BF16), with successful local runs on Apple Silicon using quantization techniques. The hosted API version, Qwen3.5-Plus, supports 1M context and integrates search and code interpreter features. This release follows other Chinese labs like Z.ai, Minimax, and Kimi in refreshing large models. The model is licensed under Apache-2.0 and is expected to be the last major release before DeepSeek v4. The news also notes Pete Steinberger joining OpenAI.
Z.ai GLM-5: New SOTA Open Weights LLM
glm-5 glm-4.5 kimi-k2.5 zhipu-ai openrouter modal deepinfra ollama qoder vercel deepseek-sparse-attention long-context model-scaling pretraining benchmarking office-productivity context-window model-deployment cost-efficiency
Zhipu AI launched GLM-5, an Opus-class model scaling from 355B to 744B parameters with DeepSeek Sparse Attention integration for cost-efficient long-context serving. GLM-5 achieves SOTA on BrowseComp and leads on Vending Bench 2, focusing on office productivity tasks and surpassing Kimi K2.5 on the GDPVal-AA benchmark. Despite broad availability on platforms like OpenRouter, Modal, DeepInfra, and Ollama Cloud, GLM-5 faces compute constraints impacting rollout and pricing. The model supports up to 200K context length and 128K max output tokens.
not much happened today
gpt-5.3-codex claude-opus-4.6 openai anthropic cursor_ai github microsoft builder-tooling cybersecurity api-access model-rollout agentic-ai long-context serving-economics throughput-latency token-efficiency workflow-design sama pierceboggan kylebrussell natolambert omarsar0 sam_altman
OpenAI launched GPT-5.3-Codex with a Super Bowl ad emphasizing "You can just build things" as a product strategy, focusing on builder tooling over chat interfaces. The model is rolling out across Cursor, VS Code, and GitHub with phased API access and is flagged as their first "high cybersecurity capability" model. Sam Altman reported over 1M Codex app downloads in the first week and strong weekly user growth. Meanwhile, Anthropic's Claude Opus 4.6 is recognized as a leading "agentic generalist" model, topping text and code leaderboards but noted for high token usage. Discussions around serving economics and "fast mode" behavior highlight practical deployment considerations. Additionally, Recursive Language Models (RLMs) introduce a novel approach using a second programmatic context space to extend long-context capabilities.
OpenAI and Anthropic go to war: Claude Opus 4.6 vs GPT 5.3 Codex
gpt-5.3-codex opus-4.6 openai anthropic nvidia agentic-coding long-context token-efficiency inference-speed hardware-software-co-design agent-platforms benchmarking software-development compiler-construction
OpenAI launched GPT-5.3-Codex, emphasizing token efficiency, inference speed, and hardware/software co-design with GB200-NVL72 and NVIDIA collaboration. The new Frontier agent platform supports business-context agents with execution environments and learning capabilities. Anthropic showcased Opus 4.6 agent teams autonomously building a clean-room C compiler booting Linux, highlighting advances in agentic coding and long-context capabilities. Community benchmarks report 2.93ร faster inference and significant efficiency gains, signaling a shift away from infinite compute budgets in 2026.
Context Graphs: Hype or actually Trillion-dollar opportunity?
glm-ocr qwen3-coder-next sera-14b zhipu-ai lmsys vllm novita-labs ollama alibaba allenai cognition cursor multimodality ocr long-context mixture-of-experts agentic-coding-models context-graphs benchmarking model-deployment model-optimization model-training jaya_gupta dharmesh_shah
Zhipu AI launched GLM-OCR, a lightweight 0.9B multimodal OCR model excelling in complex document understanding with top benchmark scores and day-0 deployment support from lmsys, vllm, and novita labs. Ollama enabled local-first usage with easy offline operation. Alibaba released Qwen3-Coder-Next, an 80B MoE model with only 3B active parameters, designed for coding agents with a massive 256K context window and trained on 800K verifiable tasks, achieving over 70% SWE-Bench Verified. The open coding ecosystem also saw Allen AI announce SERA-14B, an on-device-friendly coding model with new datasets. The emerging concept of Context Graphs was highlighted as a promising framework for data and agent traceability, with initiatives like Cursor's Agent Trace specifying context graphs for coding agents, emphasizing potential improvements in agent performance and customer-driven adoption. This coverage reflects ongoing innovation in multimodality, long-context, mixture-of-experts, and agentic coding models.
OpenEvidence, the โChatGPT for doctors,โ raises $250m at $12B valuation, 12x from $1b last Feb
claude claude-3 claude-opus gpt-5.2 gemini-3-flash-high openevidence anthropic podium openai google gemini agentic-ai model-alignment performance-evaluation memory-optimization long-context benchmarking multi-agent-systems reinforcement-learning daniel_nadler amanda_askell eric_rea tom_loverro garry_tan omarsar0 brendanfoody deredleritt3r
OpenEvidence raised $12 billion, a 12x increase from last year, with usage by 40% of U.S. physicians and over $100 million in annual revenue. Anthropic released a new Claude model constitution under CC0 1.0, framing it as a living document for alignment and training. Podium reported over $100 million ARR from 10,000+ AI agents, shifting from software sales to AI operators. Innovations in agent memory and reliability include the Agent Cognitive Compressor (ACC) and multi-agent scientific workflows via MCP-SIM. Agentic benchmarking shows challenges in long-horizon tasks with models like Gemini 3 Flash High, GPT-5.2 High, and Claude Opus 4.5 High scoring modestly on professional services and legal research benchmarks.
not much happened today
glm-4.7-flash glm-4.7 glm-4.5 qwen3-vl qwen meta-ai-fair carnegie-mellon sakana-ai zhipu-ai transformer-memory model-architecture mixture-of-experts adaptive-position-encoding long-context model-compression inference-optimization local-inference model-deployment benchmarking coding agentic-ai
AI News for 1/16/2026-1/19/2026 covers new architectures for scaling Transformer memory and context, including STEM from Carnegie Mellon and Meta AI, which replaces part of the FFN with a token-indexed embedding lookup enabling CPU offload and asynchronous prefetch. RePo from Sakana AI introduces adaptive positional reordering to improve robustness on noisy and long-range contexts. Model releases highlight Zhipu AI's GLM-4.7-Flash, a 30B-class MLA + small MoE model optimized for coding and agentic tasks, noted for strong benchmark performance and a compression narrative from larger to smaller models. Inference and deployment updates include mlx-lm 0.30.3 supporting GLM-4.7-Flash with efficient 4-bit performance on laptops. The report emphasizes practical takeaways on static sparsity, adaptive ordering, and the resurgence of small, fast models for interactive tasks. "Sparse capacity doesnโt have to mean MoE routers + expert parallelism; static sparsity can be systems-friendly."
Apple picks Google's Gemini to power Siri's next generation
gemini claude chatgpt engram apple google openai anthropic deepseek conditional-memory long-context hashing memory-optimization transformers model-scaling sparsity hardware-optimization model-architecture ai-healthcare model-optimization
Apple has decided to power Siri with Google's Gemini models and cloud technology, marking a significant partnership and a setback for OpenAI, which was initially partnered with Apple. Anthropic launched "Cowork," a product preview for Claude's coding capabilities, sparking discussions about "LLM OS". OpenAI introduced ChatGPT Health and acquired Torch to expand in healthcare AI. DeepSeek unveiled Engram, a new conditional memory module that enables O(1) lookup-style memory for static patterns, improving long-context handling and offering hardware-friendly optimizations to scale knowledge capacity efficiently. Engram is positioned as a key modeling primitive for next-gen sparse models, with ongoing community debate about its architectural merits and practical impact.
not much happened today
qwen-image-layered kling-2.6 gwm-1 gen-4.5 gemini-3-flash gpt-5.2 codex-cli opus-4.5 alibaba kling-ai runway google anthropic openai image-decomposition motion-control video-generation agentic-reinforcement-learning long-context model-degradation benchmarking tool-use prompt-engineering ankesh_anand
Alibaba released Qwen-Image-Layered, an open-source model enabling Photoshop-grade layered image decomposition with recursive infinite layers and prompt-controlled structure. Kling 2.6 introduced advanced motion control for image-to-video workflows, supported by a creator contest and prompt recipes. Runway unveiled the GWM-1 family with frame-by-frame video generation and Gen-4.5 updates adding audio and multi-shot editing. In LLM platforms, Gemini 3 Flash leads benchmarks over GPT-5.2, attributed to agentic reinforcement learning improvements post-distillation. Users note GPT-5.2 excels at long-context tasks (~256k tokens) but face UX limitations pushing some to use Codex CLI. Discussions around Anthropic Opus 4.5 suggest perceived model degradation linked to user expectations.
Claude Skills grows: Open Standard, Directory, Org Admin
claude-skills gpt-5.2-codex gemini-3-flash functiongemma t5gemma-2 anthropic openai google-deepmind hugging-face agentic-ai fine-tuning long-context tool-calling on-device-ai multimodality security workflow-optimization sama gregbrockman philschmid
Claude Skills are gaining significant traction since their launch in October, with a milestone of 100k views in one day for the Claude Skills talk, signaling growing adoption and importance. Announcements include org admin support, a new Skills Directory, and the move to an open standard named Agent Skills. In frontier model launches, OpenAI released GPT-5.2-Codex, touted as the best agentic coding model with improvements in native compaction, long-context reliability, and tool-calling, emphasizing real-world security impacts. Google DeepMind introduced Gemini 3 Flash, focusing on speed as a product feature impacting workflows and user engagement, alongside FunctionGemma and T5Gemma 2, emphasizing on-device deployment, fine-tuning, and multimodality.
OpenAI GPT Image-1.5 claims to beat Nano Banana Pro, #1 across all Arenas, but completely fails Vibe Checks
gpt-image-1.5 nano-banana-pro mimo-v2-flash deepseek-v3.2 openai gemini xiaomi lmsys deepseek openrouter image-generation instruction-following benchmarking model-efficiency long-context multi-token-prediction hybrid-attention model-optimization inference-speed agentic-workflows model-architecture model-quantization fuli_luo eliebakouch
OpenAI released its new image model GPT Image 1.5, featuring precise image editing, better instruction following, improved text and markdown rendering, and faster generation up to 4ร. Despite topping multiple leaderboards like LMArena (1277), Design Arena (1344), and AA Arena (1272), user feedback from Twitter, Reddit, and Discord communities is largely negative compared to Nano Banana Pro by Gemini. Xiaomi introduced the MiMo-V2-Flash, a 309B MoE model optimized for inference efficiency with 256K context window, achieving state-of-the-art scores on SWE-Bench. The model uses Hybrid Sliding Window Attention and multi-token prediction, offering significant speedups and efficiency improvements. The timing of OpenAI's launch amid competition from Gemini and Nano Banana Pro affects user sentiment, highlighting challenges in benchmarking relevance.
NVIDIA Nemotron 3: hybrid Mamba-Transformer completely open source models from 30B to 500B
nemotron-3-nano qwen3-30b-a3b-base nvidia huggingface togethercompute baseten vllm llamaindex hybrid-architecture mixture-of-experts reinforcement-learning long-context model-release open-source-models model-training model-optimization benchmarking agent-training ctnzr andrew_n_carr awnihannun
NVIDIA has released Nemotron 3 Nano, a fully open-source hybrid Mamba-Transformer Mixture-of-Experts (MoE) model with a 30B parameter size and a 1 million token context window. It includes open weights, training recipes, datasets, and an RL environment suite called NeMo Gym, supporting commercial use under the NVIDIA Open Model License. The model achieves state-of-the-art results on benchmarks like SWE-Bench and Artificial Analysis Intelligence Index, outperforming Qwen3-30B A3B. Ecosystem support is immediate with integrations into inference stacks like vLLM, llama.cpp, and Baseten. Upcoming larger models, Nemotron Super and Ultra, will feature NVFP4 pretraining and LatentMoE routing to optimize compute. This release marks a significant milestone for open-source American AI with comprehensive open assets and advanced hybrid architecture.
not much happened today
gpt-5.2 opus-4.5 gemini-3-pro gpt-5.1 olmo-3.1-32b qwen3-vl-235b openai allen_ai mistral-ai ollama lmstudio thinkymachines reinforcement-learning model-benchmarking long-context model-quantization model-optimization inference-speed sparsity fine-tuning vision sama scaling01 akhaliq artificialanlys lechmazur acerfur epochairesearch
GPT-5.2 shows mixed performance in public evaluations, excelling in agentic tasks but at a significantly higher cost (~$620/run) compared to Opus 4.5 and GPT-5.1. It performs variably on reasoning and coding benchmarks, with some improvements on long-context tasks. Extended "reasoning effort" settings notably impact results. Aggregators rank Gemini 3 Pro above GPT-5.2 in task persistence. OpenAI released sparse activation models sparking debate on sparsity vs MoE architectures. Allen AI's Olmo 3.1 (32B) advances open reinforcement learning scale with substantial compute investment (~125k H100 hours). Mistral's Devstral-2 and llama.cpp improve local inference infrastructure with new features like GGUF support and distributed speedups. Tinker platform goes GA with vision input and finetuning support for Qwen3-VL-235B.
GPT-5.2 (Instant/Thinking/Pro): 74% on GDPVal, 1.4x cost of GPT 5.1, on 10 Year OpenAI Anniversary
gpt-5.2 openai scientific-reasoning knowledge-work long-context benchmarking performance-optimization pricing software-engineering vision sama yanndubs polynoamial scaling01
OpenAI celebrates its 10 year anniversary with the launch of GPT-5.2, featuring significant across-the-board improvements including a rare 40% price increase. GPT-5.2 shows strong performance gains in scientific reasoning, knowledge work, and economic value tasks, achieving over 70.9% human expert parity on GDPval tasks and reaching 90.5% on ARC-AGI-1 with a large efficiency gain. Despite some mixed results in coding benchmarks and vision capabilities, GPT-5.2 is well received as a major update with extended context and tiered reasoning controls. Pricing is set at $1.75/M input and $14/M output tokens with a 90% cache discount. The update is live in ChatGPT and API, marking a significant milestone for OpenAI's LLM development.
not much happened today
glm-4.6v glm-4.6v-flash jina-vlm-2b hugging-face zhipu-ai jina-ai google-deepmind axiomprover fine-tuning multimodality model-optimization long-context mechanistic-interpretability formal-methods sequence-architectures reinforcement-learning lioronai akshay_pachaar _akhaliq ben_burtenshaw vllm_project prince_canuma zenmuxai eliebakouch theturingpost axiommathai neelnanda5 sarahookr
Claude Code Skills gains attention with a published talk and Hugging Face's new "skill" enabling one-line fine-tuning pipelines for models from ~0.5B to 70B parameters, supporting SFT, DPO, and GRPO, costing as low as ~$0.30 for small runs. Zhipu AI launches multimodal models GLM-4.6V (106B params MoE) and GLM-4.6V-Flash (9B dense), featuring 128k context and native multimodal function calling, with free Flash variant and API pricing detailed. Jina AI releases Jina-VLM (2B), a compact multilingual VLM excelling in diagrams and documents with top benchmark scores. At NeurIPS 2025, research highlights include Google's post-Transformer sequence architectures (Moneta, Yaad, Memora) showing up to 20% gains in long-context retrieval, AxiomProver's autonomous Lean system solving 9/12 Putnam 2025 problems rapidly, and mechanistic interpretability advances discussed by Chris Olah emphasizing scalable tooling.
OpenRouter's State of AI - An Empirical 100 Trillion Token Study
grok-code-fast gemini-3 gemini-3-deep-think gpt-5.1-codex-max openrouter deepseek anthropic google google-deepmind reasoning coding tokenization long-context model-architecture benchmarking agentic-ai prompt-engineering quocleix noamshazeer mirrokni
OpenRouter released its first survey showing usage trends with 7 trillion tokens proxied weekly, highlighting a 52% roleplay bias. Deepseek's open model market share has sharply declined due to rising coding model usage. Reasoning model token usage surged from 0% to over 50%. Grok Code Fast shows high usage, while Anthropic leads in tool calling and coding requests with around 60% share. Input tokens quadrupled and output tokens tripled this year, driven mainly by programming use cases, which dominate spending and volume. Google launched Gemini 3 Deep Think, featuring parallel thinking and achieving 45.1% on ARC-AGI-2 benchmarks, and previewed Titans, a long-context neural memory architecture scaling beyond 2 million tokens. These advances were shared by Google DeepMind and Google AI on Twitter.
Mistral 3: Mistral Large 3 + Ministral 3B/8B/14B open weights models
mistral-large-3 ministral-3 clara-7b-instruct gen-4.5 claude-code mistral-ai anthropic apple runway moondream sparse-moe multimodality benchmarking open-source model-licensing model-performance long-context inference-optimization instruction-following local-inference code-generation model-integration anjney_midha _akhaliq alexalbert__ _catwu mikeyk
Mistral has launched the Mistral 3 family including Ministral 3 models (3B/8B/14B) and Mistral Large 3, a sparse MoE model with 675B total parameters and 256k context window, all under an Apache 2.0 open license. Early benchmarks rank Mistral Large 3 at #6 among open models with strong coding performance. The launch includes broad ecosystem support such as vLLM, llama.cpp, Ollama, and LM Studio integrations. Meanwhile, Anthropic acquired the open-source Bun runtime to accelerate Claude Code, which reportedly reached a $1B run-rate in ~6 months. Anthropic also announced discounted Claude plans for nonprofits and shared insights on AI's impact on work internally.
not much happened today
claude-opus-4.5 qwen-3-4b qwen-3-8b qwen-3-14b deepseek-r1 anthropic booking.com perplexity-ai langchain claude scaling01 deepseek qwen prefect agent-systems multi-agent-systems reasoning benchmarking cost-efficiency model-optimization long-context memory-management reinforcement-learning model-performance multi-agent-communication latent-representation inference-cost software-integration jeremyphoward alexalbert__ omarsar0 lingyang_pu dair_ai
Anthropic introduces durable agents and MCP tasks for long-running workflows, with practical engineering patterns and integrations like Prefect. Booking.com deploys a large-scale agent system improving customer satisfaction using LangGraph, Kubernetes, GPT-4 Mini, and Weaviate. Perplexity rolls out user-level memory and virtual try-on features. Claude Opus 4.5 leads on LisanBench and Code Arena WebDev benchmarks with mixed community feedback on its "thinking" and "non-thinking" modes, while improving cost-efficiency and UX with batch APIs and context compaction. Research on multi-agent systems shows LatentMAS reduces communication tokens by 70-84% and improves accuracy using Qwen3 models, and reasoning trace distillation achieves significant token reduction with maintained accuracy, highlighting the importance of reasoning trace style.
not much happened today
qwen3-max-thinking minimax-m2 claude-3-sonnet llamaindex-light chronos-2 openai aws microsoft nvidia gpu_mode vllm alibaba arena llamaindex amazon anthropic gradio compute-deals gpu-optimization kernel-optimization local-serving reasoning long-context benchmarks long-term-memory time-series-forecasting agent-frameworks oauth-integration developer-tools sama gdb andrewcurran_ a1zhang m_sirovatka omarsar0 _philschmid
OpenAI and AWS announced a strategic partnership involving a $38B compute deal to deploy hundreds of thousands of NVIDIA GB200 and GB300 chips, while Microsoft secured a license to ship NVIDIA GPUs to the UAE with a planned $7.9B datacenter investment. A 3-month NVFP4 kernel optimization competition on Blackwell B200s was launched by NVIDIA and GPU_MODE with prizes including DGX Spark and RTX 50XX GPUs. vLLM gains traction for local LLM serving, exemplified by PewDiePie's adoption. Alibaba previewed the Qwen3-Max-Thinking model hitting 100% on AIME 2025 and HMMT benchmarks, signaling advances in reasoning with tool use. The MIT-licensed MiniMax-M2 230B MoE model topped the Arena WebDev leaderboard, tying with Claude Sonnet 4.5 Thinking 32k. Critiques emerged on OSWorld benchmark stability and task validity. LlamaIndex's LIGHT framework demonstrated significant improvements in long-term memory tasks over raw context and RAG baselines, with gains up to +160.6% in summarization at 10M tokens. Amazon introduced Chronos-2, a time-series foundation model for zero-shot forecasting. The MCP ecosystem expanded with new tools like mcp2py OAuth integration and Gemini Docs MCP server, alongside a build sprint by Anthropic and Gradio offering substantial credits and prizes. "OSWorld doesnโt really existโdifferent prompt sets = incomparable scores" highlights benchmarking challenges.
not much happened today
Poolside raised $1B at a $12B valuation. Eric Zelikman raised $1B after leaving Xai. Weavy joined Figma. New research highlights FP16 precision reduces training-inference mismatch in reinforcement-learning fine-tuning compared to BF16. Kimi AI introduced a hybrid KDA (Kimi Delta Attention) architecture improving long-context throughput and RL stability, alongside a new Kimi CLI for coding with agent protocol support. OpenAI previewed Agent Mode in ChatGPT enabling autonomous research and planning during browsing.
not much happened today
kimi-linear kimi-delta-attention minimax-m2 looped-llms aardvark-gpt-5 moonshot-ai minimax bytedance princeton mila openai cursor cognition hkust long-context attention-mechanisms agentic-ai tool-use adaptive-compute coding-agents performance-optimization memory-optimization reinforcement-learning model-architecture kimi_moonshot scaling01 uniartisan omarsar0 aicodeking songlinyang4 iscienceluvr nrehiew_ gdb embeddedsec auchenberg simonw
Moonshot AI released Kimi Linear (KDA) with day-0 infrastructure and strong long-context metrics, achieving up to 75% KV cache reduction and 6x decoding throughput. MiniMax M2 pivoted to full attention for multi-hop reasoning, maintaining strong agentic coding performance with 200k context and ~100 TPS. ByteDance, Princeton, and Mila introduced Looped LLMs showing efficiency gains comparable to larger transformers. OpenAI's Aardvark (GPT-5) entered private beta as an agentic security researcher for scalable vulnerability discovery. Cursor launched faster cloud coding agents, though transparency concerns arose regarding base-model provenance. Cognition released a public beta for a desktop/mobile tool-use agent named Devin. The community discussed advanced attention mechanisms and adaptive compute techniques.
DeepSeek-OCR finds vision models can decode 10x more efficiently with ~97% accuracy of text-only, 33/200k pages/day/A100
deepseek-ocr deepseek3b-moe-a570m veo-3.1 deepseek-ai google-deepmind krea ocr vision multimodality model-compression long-context model-architecture video-generation autoregressive-models model-efficiency precision-editing karpathy teortaxestex reach_vb _akhaliq eliebakouch vikhyatk demishassabis
As ICCV 2025 begins, DeepSeek releases a novel DeepSeek-OCR 3B MoE vision-language model that compresses long text as visual context with high accuracy and efficiency, challenging traditional tokenization approaches. The model achieves ~97% decoding precision at <10ร compression and processes up to ~33M pages/day on 20 A100-40G nodes, outperforming benchmarks like GOT-OCR2.0. Discussions highlight the potential for unlimited context windows and tokenization-free inputs, with contributions from @karpathy, @teortaxesTex, and others. In video generation, google-deepmind's Veo 3.1 leads community benchmarks with advanced precision editing and scene blending, while Krea open-sources a 14B autoregressive video model enabling realtime long-form generation at ~11 FPS on a single B200 GPU.
The Karpathy-Dwarkesh Interview delays AGI timelines
claude-haiku-4.5 gpt-5 arch-router-1.5b anthropic openai huggingface langchain llamaindex google epoch-ai reasoning long-context sampling benchmarking data-quality agent-frameworks modular-workflows ide-extensions model-routing graph-first-agents real-world-grounding karpathy aakaran31 du_yilun giffmana omarsar0 jeremyphoward claude_code mikeyk alexalbert__ clementdelangue jerryjliu0
The recent AI news highlights the Karpathy interview as a major event, alongside significant discussions on reasoning improvements without reinforcement learning, with test-time sampling achieving GRPO-level performance. Critiques on context window marketing reveal effective limits near 64K tokens, with Claude Haiku 4.5 showing competitive reasoning speed. GPT-5 struggles with advanced math benchmarks, and data quality issues termed "Brain Rot" affect model reasoning and safety. In agent frameworks, Anthropic Skills enable modular coding workflows, OpenAI Codex IDE extensions enhance developer productivity, and HuggingChat Omni introduces meta-routing across 100+ open models using Arch-Router-1.5B. LangChain and LlamaIndex advance graph-first agent infrastructure, while Google Gemini integrates with Google Maps for real-world grounding.
Claude Agent Skills - glorified AGENTS.md? or MCP killer?
claude-4.5-haiku claude chatgpt huggingchat-omni anthropic openai microsoft perplexity-ai huggingface groq cerebras togethercompute agent-skills document-processing long-context reasoning multi-model-routing memory-management voice vision simonwillison alexalbert__ mustafasuleyman yusuf_i_mehdi aravsrinivas
Anthropic achieves a rare feat with back-to-back AI news headlines featuring Claude's new Skillsโa novel way to build specialized agents using Markdown files, scripts, and metadata to handle tasks like creating and reading PDFs, Docs, and PPTs. Simon Willison calls this a "bigger deal than MCP," predicting a "Cambrian explosion in Skills." Meanwhile, Anthropic launches Claude 4.5 Haiku with strong reasoning and long-context capabilities, priced competitively. Other updates include OpenAI's ChatGPT memory management improvements, Windows 11 Copilot voice and vision features, and HuggingChat Omni routing across 115 open-source models from 15 providers. These developments highlight advances in agent skills, document processing, long-context reasoning, and multi-model routing.
not much happened today
7m-tiny-recursive-model jamba-reasoning-3b qwen3-omni qwen-image-edit-2509 colbert-nano agentflow samsung lecuun ai21-labs alibaba coreweave weights-biases openpipe stanford recursive-reasoning density-estimation multimodality long-context retrieval serverless-reinforcement-learning agentic-systems model-efficiency reinforcement-learning transformers rasbt jm_alexia jiqizhixin randall_balestr corbtt shawnup _akhaliq
Samsung's 7M Tiny Recursive Model (TRM) achieves superior reasoning on ARC-AGI and Sudoku with fewer layers and MLP replacing self-attention. LeCun's team introduces JEPA-SCORE, enabling density estimation from encoders without retraining. AI21 Labs releases Jamba Reasoning 3B, a fast hybrid SSM-Transformer model supporting up to 64K context tokens. Alibaba's Qwen3 Omni/Omni Realtime offers a unified audio-video-text model with extensive language and speech support, outperforming Gemini 2.0 Flash on BigBench Audio. Alibaba also debuts Qwen Image Edit 2509, a top open-weight multi-image editing model. ColBERT Nano models demonstrate effective retrieval at micro-scale parameter sizes. In reinforcement learning, CoreWeave, Weights & Biases, and OpenPipe launch serverless RL infrastructure reducing costs and speeding training. Stanford's AgentFlow presents an in-the-flow RL system with a 7B backbone outperforming larger models on agentic tasks. This update highlights advances in recursive reasoning, density estimation, multimodal architectures, long-context modeling, retrieval, and serverless reinforcement learning.
Anthropic Claude Sonnet 4.5, Claude Code 2.0, new VS Code Extensions
claude-sonnet-4.5 claude-code-v2 deepseek-v3.2-exp anthropic deepseek openai stripe swe-bench finance law stem code-execution context-editing memory-management api chrome-extension generative-ui sparse-attention long-context cost-efficiency john_schulman mike_krieger
Anthropic launched a major update with Claude Sonnet 4.5, achieving 77.2% SWE-Bench verified performance and improvements in finance, law, and STEM. They also released Claude Code v2 featuring checkpoints, a refreshed terminal, and a native VS Code extension, plus a new mascot Clawd. The Claude API gained context editing and memory tools, and the Claude Agent SDK was introduced. The Claude.ai apps now support code execution and file creation, with a Chrome extension available for Max users. Additionally, Imagine with Claude offers a generative UI research preview. Reception has been positive from developers and third-party evaluators. Meanwhile, DeepSeek released V3.2-Exp with a new Sparse Attention algorithm, significantly reducing long-context costs and cutting API prices by over 50%, while maintaining quality.
GDPVal finding: Claude Opus 4.1 within 95% of AGI (human experts in top 44 white collar jobs)
claude-4.1-opus gpt-5-high gptnext gemini-2.5-flash gemini-2.5-flash-lite deepseek-v3.1-terminus google-chirp-2 qwen-2.5b openai anthropic google nvidia artificial-analysis deepseek benchmarking agentic-ai tool-use long-context speech-to-text model-evaluation reasoning pricing model-performance kevinweil gdb dejavucoder yuchenj_uw lhsummers
OpenAI's Evals team released GDPval, a comprehensive evaluation benchmark covering 1,320 tasks across 44 predominantly digital occupations, assessing AI models against human experts with 14 years average experience. Early results show Claude 4.1 Opus outperforming human experts in most categories and GPT-5 high trailing behind, with projections that GPTnext could match human performance by mid-2026. The benchmark is positioned as a key metric for policymakers and labor impact forecasting. Additionally, Artificial Analysis reported improvements in Gemini 2.5 Flash/Flash-Lite and DeepSeek V3.1 Terminus models, alongside new speech-to-text benchmarks (AA-WER) highlighting leaders like Google Chirp 2 and NVIDIA Canary Qwen2.5B. Agentic AI advances include Kimi OK Computer, an OS-like agent with extended tool capabilities and new vendor verification tools.
not much happened today
gpt-5-codex vllm-0.10.2 qwen3-next-80b hunyuanimage-2.1 openai microsoft perplexity-ai huggingface amd tencent lmstudio agentic-ai ide context-windows inference distributed-inference reinforcement-learning robotics long-context model-optimization text-to-image multimodality model-licenses gdb teknium1 finbarrtimbers thsottiaux theturingpost pierceboggan amandaksilver aravsrinivas sergiopaniego art_zucker danielhanchen rwojo awnihannun
GPT-5 Codex rollout shows strong agentic coding capabilities with some token bloat issues. IDEs like VS Code Insiders and Cursor 1.6 enhance context windows and model integration. vLLM 0.10.2 supports aarch64 and NVIDIA GB200 with performance improvements. AMD ROCm updates add modern attention, sparse MoE, and distributed inference. TRL introduces Context Parallelism for long-context training. Robotics and RL data pipelines improve with Unsloth and LeRobotDataset v3. Qwen3-Next-80B runs efficiently on Mac M4 Max with MLX. Tencent's HunyuanImage 2.1 is a 17B bilingual text-to-image model with 2048ร2048 resolution and restricted open weights.
GPT-5 Codex launch and OpenAI's quiet rise in Agentic Coding
gpt-5-codex qwen3-next-80b openai alibaba together-ai nvidia agentic-ai software-engineering long-context mixture-of-experts model-optimization cuda-acceleration inference-efficiency routing task-adaptive-thinking sama swyx omarsar0 ofirpress
OpenAI released GPT-5-Codex, an agentic coding model optimized for long-running software engineering tasks with dynamic task-adaptive thinking, multi-hour autonomy, and improved code quality. It achieves 51% accuracy on an unreleased large refactor benchmark and integrates deeply with developer tools like Xcode. Meanwhile, Alibaba launched Qwen3-Next-80B, a hybrid MoE model with native long-context support (262k tokens, extensible to 1M+), targeting efficient reasoning and repository-scale code analysis, supported by Together AI and NVIDIA with CUDA-accelerated attention. The trend towards hybrid SSM + MoE architectures is noted, emphasizing efficiency and scaling in China and US training regimes. Community discussions highlight the importance of variable compute and routing for inference efficiency and quality.
not much happened today
mobilellm-r1 qwen3-next-80b-a3b gpt-5 meta-ai-fair huggingface alibaba openai reasoning model-efficiency hybrid-attention long-context benchmarking agent-evaluation hallucination-detection model-calibration inference-complexity model-pricing _akhaliq tacocohen pkirgis sayashk
Meta released MobileLLM-R1, a sub-1B parameter reasoning model family on Hugging Face with strong small-model math accuracy, trained on 4.2T tokens. Alibaba introduced Qwen3-Next-80B-A3B with hybrid attention, 256k context window, and improved long-horizon memory, priced competitively on Alibaba Cloud. Meta AI FAIR fixed a benchmark bug in SWE-Bench affecting agent evaluation. LiveMCP-101 benchmark shows frontier models like GPT-5 underperform on complex tasks with common failure modes cataloged. OpenAI highlights hallucination issues due to benchmark incentives, proposing calibration improvements. Community demos and tooling updates continue to evolve.
Oracle jumps +36% in a day after winning $300B OpenAI contract
qwen3-235b qwen3-4b qwen2.5-7b vllm oracle openai microsoft moonshot-ai vllm-project thinking-machines-lab meta reinforcement-learning model-weight-updates deterministic-inference benchmarking long-context model-optimization cuda distributed-training kimi_moonshot arankomatsuzaki qgallouedec cHHillee woosuk_k stasbekman
Oracle's OCI division reported a stunning +359% revenue bookings growth to $455B with cloud revenue guidance of $144B by 2030, driven significantly by a large deal with OpenAI amid tensions with Microsoft. On AI infrastructure, Moonshot AI released Kimiโs checkpoint-engine, enabling rapid weight updates on 1T-parameter models across thousands of GPUs, integrating with vLLM. RLFactory introduced a plug-and-play reinforcement learning framework for tool-using agents, showing smaller models outperforming larger ones. TRL v0.23 added context parallelism for long-context training. Thinking Machines Lab published research on deterministic inference pipelines, making vLLM deterministic for Qwen models. Meta launched BackendBench, a PyTorch benchmarking tool.
Kimi K2โ0905 and Qwen3โMax preview: two 1T open weights models launched
kimi-k2-0905 qwen-3-max qwen-3 moonshot-ai alibaba huggingface together-ai groq lmsys openrouter llamaindex long-context agents coding tool-use model-evaluation instruction-following context-windows semantic-search discriminator-models swyx karpathy willdepue levie bebischof andrew_n_carr bigeagle_xd
Moonshot AI updated their Kimi K2-0905 open model with doubled context length to 256k tokens, improved coding and tool-calling, and integration with agent scaffolds. Alibaba released Qwen 3 Max, a 1 trillion parameter model with agent-oriented behavior, available via Qwen Chat, Alibaba Cloud API, and OpenRouter. The community highlights China's dominance in open models and debates around meaningful evaluation methods for code agents, emphasizing long-horizon and domain-specific evals. Influential voices like @swyx and @karpathy discuss the importance of practical evals and discriminator models for ranking outputs.
Cohere Command A Reasoning beats GPT-OSS-120B and DeepSeek R1 0528
command-a-reasoning deepseek-v3.1 cohere deepseek intel huggingface baseten vllm-project chutes-ai anycoder agentic-ai hybrid-models long-context fp8-training mixture-of-experts benchmarking quantization reasoning coding-workflows model-pricing artificialanlys reach_vb scaling01 cline ben_burtenshaw haihaoshen jon_durbin _akhaliq willccbb teortaxestex
Cohere's Command A Reasoning model outperforms GPT-OSS in open deep research capabilities, emphasizing agentic use cases for 2025. DeepSeek-V3.1 introduces a hybrid reasoning architecture toggling between reasoning and non-reasoning modes, optimized for agentic workflows and coding, with extensive long-context pretraining (~630B tokens for 32k context, ~209B for 128k), FP8 training, and a large MoE expert count (~37B). Benchmarks show competitive performance with notable improvements in SWE-Bench and other reasoning tasks. The model supports a $0.56/M input and $1.68/M output pricing on the DeepSeek API and enjoys rapid ecosystem integration including HF weights, INT4 quantization by Intel, and vLLM reasoning toggles. Community feedback highlights the hybrid design's pragmatic approach to agent and software engineering workflows, though some note the lack of tool use in reasoning mode.
DeepSeek V3.1: 840B token continued pretrain, beating Claude 4 Sonnet at 11% of its cost
deepseek-v3.1 seed-oss-36b computerrl gemini-2.5-pro gpt-5 claude-code gpt-oss-120b gpt-oss-20b deepseek bytedance zhipu-ai github microsoft anthropic together-ai baseten huggingface token-efficiency coding agentic-benchmarks long-context reinforcement-learning developer-tools fine-tuning multinode-training model-release teortaxestex rasbt lukehoban burkeholland _catwu cline winglian
DeepSeek released DeepSeek V3.1, a quietly rolled out open model with an 128K context window and improvements in token efficiency, coding, and agentic benchmarks. ByteDance launched the permissive Seed-OSS 36B model on Hugging Face, noted for long-context and reasoning capabilities. Zhipu AI introduced ComputerRL, a reinforcement learning framework for computer-use agents, achieving strong benchmark results. In developer tooling, GitHub Copilot expanded globally, Microsoft VS Code integrated Gemini 2.5 Pro and updated GPT-5 agent prompts, and Anthropic launched Claude Code seats with spend controls. Open-source fine-tuning advances include Together AI adding SFT for gpt-oss-120B/20B and Baseten enabling multinode 120B training with Truss CLI. The community noted mixed performance and ongoing post-training adjustments for DeepSeek V3.1.
not much happened today
gpt-5 gpt-5-mini gpt-5-nano claude-sonnet-4 glm-4.5v genie-3 gemini-app qwen-image-distilled matrix-game-2.0 jan-v1 qwen3-4b-thinking openai anthropic zhipu-ai google-deepmind alibaba skywork jan-ai context-window multimodality reinforcement-learning agentic-tasks video-generation image-generation real-time-systems web-search model-accuracy developer-tools open-source-models long-context model-scaling
OpenAI released the GPT-5 series including GPT-5-mini and GPT-5-nano, with mixed user feedback on performance and API behavior. Anthropic extended Claude Sonnet 4 context window to 1 million tokens, a 5x increase, enhancing large document processing. Zhipu AI launched the open-source multimodal GLM-4.5V model with improvements in RL scaling and agentic tasks. Google DeepMind showcased the video generation model Genie 3 and updated the Gemini App with new features like Deep Think and Gemini Live. Alibaba Qwen released the distilled image model Qwen-Image distilled and enhanced their Deep Research capabilities. Open source models like Skywork's Matrix-Game 2.0 and Jan.ai's Jan-v1 (built on Qwen3-4B-Thinking) were introduced, focusing on real-time world modeling and web search respectively. Developer tools such as Claude Code and Cursor were also highlighted.
OpenAI rolls out GPT-5 and GPT-5 Thinking to >1B users worldwide; -mini and -nano help claim Pareto Frontier
gpt-5 gpt-5-mini gpt-5-nano claude-4.1-sonnet claude-4.1-opus openai cursor_ai jetbrains microsoft notion perplexity_ai factoryai model-architecture context-windows pricing-models coding long-context prompt-engineering model-benchmarking model-integration tool-use reasoning sama scaling01 jeffintime embirico mustafasuleyman cline lmarena_ai nrehiew_ ofirpress sauers_
OpenAI launched GPT-5, a unified system featuring a fast main model and a deeper thinking model with a real-time router, supporting up to 400K context length and aggressive pricing that reclaims the Pareto Frontier of Intelligence. The rollout includes variants like gpt-5-mini and gpt-5-nano with significant cost reductions, and integrations with products such as ChatGPT, Cursor AI, JetBrains AI Assistant, Microsoft Copilot, Notion AI, and Perplexity AI. Benchmarks show GPT-5 performing strongly in coding and long-context reasoning, roughly matching Claude 4.1 Sonnet/Opus on SWE-bench Verified. The launch was accompanied by a GPT-5 prompting cookbook and notable community discussions on pricing and performance.
not much happened today
kimi-k2 qwen3-235b-a22b qwen3-coder-480b-a35b gemini-2.5-flash-lite mistral-7b deepseek-v3 moonshot-ai alibaba google google-deepmind openai hugging-face vllm-project mixture-of-experts agentic-ai model-optimization model-training benchmarking code-generation long-context multimodality math reinforcement-learning model-architecture model-performance open-source alignment demishassabis rasbt alexwei_ yitayml
Moonshot AI released the Kimi K2, a 1-trillion parameter ultra-sparse Mixture-of-Experts (MoE) model with the MuonClip optimizer and a large-scale agentic data pipeline using over 20,000 tools. Shortly after, Alibaba updated its Qwen3 model with the Qwen3-235B-A22B variant, which outperforms Kimi K2 and other top models on benchmarks like GPQA and AIME despite being 4.25x smaller. Alibaba also released Qwen3-Coder-480B-A35B, a MoE model specialized for coding with a 1 million token context window. Google DeepMind launched Gemini 2.5 Flash-Lite, a faster and more cost-efficient model outperforming previous versions in coding, math, and multimodal tasks. The MoE architecture is becoming mainstream, with models like Mistral, DeepSeek, and Kimi K2 leading the trend. In mathematics, an advanced Gemini model achieved a gold medal level score at the International Mathematical Olympiad (IMO), marking a first for AI. An OpenAI researcher noted their IMO model "knew" when it did not have a correct solution, highlighting advances in model reasoning and self-awareness.
ChatGPT Agent: new o* model + unified Deep Research browser + Operator computer use + Code Interpreter terminal
o3 o4 gptnext openai reinforcement-learning benchmarking model-performance model-risk long-context model-deployment fine-tuning sama gdb kevinweil xikun_zhang_ keren_gu boazbaraktcs
OpenAI launched the ChatGPT Agent, a new advanced AI system capable of browsing the web, coding, analyzing data, and creating reports, marking a significant step towards human-like computer use. The agent, distinct from and superior to o3, is considered the first public exposure of what was internally called o4, now merged into GPTNext. It features end-to-end reinforcement learning, can operate for extended periods (tested up to 2 hours), and is classified as "High" risk for biological misuse, with safeguards activated. Early benchmarks show mixed results, excelling in some tests like WebArena and BrowserComp but underperforming on others like PaperBench. Key figures involved include Sam Altman, Greg Brockman, and Kevin Weil, with technical insights from xikun_zhang_ and risk commentary from KerenGu and boazbaraktcs. The launch sparked speculation about GPT-5, which was confirmed not to be the case.
Voxtral - Mistral's SOTA ASR model in 3B (mini) and 24B ("small") sizes beats OpenAI Whisper large-v3
voxtal-3b voxtal-24b kimi-k2 mistral-ai moonshot-ai groq together-ai deepinfra huggingface langchain transcription long-context function-calling multilingual-models mixture-of-experts inference-speed developer-tools model-integration jeremyphoward teortaxestex scaling01 zacharynado jonathanross321 reach_vb philschmid
Mistral surprises with the release of Voxtral, a transcription model outperforming Whisper large-v3, GPT-4o mini Transcribe, and Gemini 2.5 Flash. Voxtral models (3B and 24B) support 32k token context length, handle audios up to 30-40 minutes, offer built-in Q&A and summarization, are multilingual, and enable function-calling from voice commands, powered by the Mistral Small 3.1 language model backbone. Meanwhile, Moonshot AI's Kimi K2, a non-reasoning Mixture of Experts (MoE) model built by a team of around 200 people, gains attention for blazing-fast inference on Groq hardware, broad platform availability including Together AI and DeepInfra, and local running on M4 Max 128GB Mac. Developer tool integrations include LangChain and Hugging Face support, highlighting Kimi K2's strong tool use capabilities.
Kimi K2 - SOTA Open MoE proves that Muon can scale to 15T tokens/1T params
kimi-k2 kimi-k2-1t deepseek-v3 grok-4 devstral-2507 gpt-4.1 sonnet-4 moonshot-ai alibaba tencent deepseek x-ai mistral-ai weights-biases hugging-face mixture-of-experts model-training model-optimization optimizer benchmarking long-context model-performance open-weights model-release yuchenj_uw andrew_n_carr scaling01 novita_labs teknium1 aravsrinivas mparakhin simonw
Moonshot AI has released Kimi K2, a 1 trillion parameter Mixture-of-Experts model trained on 15.5 trillion tokens using the new MuonClip optimizer, achieving state-of-the-art results on benchmarks like SWE-Bench Verified (65.8%) and TAU2 (58.4%). This model is competitive with GPT-4.1 and Sonnet 4 on non-thinking tasks and is available under an MIT license. Meanwhile, xAI announced Grok-4, noted for its "LEAST censored frontier model" status and strong long-context performance but criticized for rushed post-training. Mistral AI updated its Devstral 2507 models with improved performance and cost efficiency. The community is excited about the potential of the MuonClip optimizer, which may surpass the long-standing AdamW optimizer in machine learning.
Grok 4: xAI succeeds in going from 0 to new SOTA LLM in 2 years
grok-4 grok-4-heavy claude-4-opus xai perplexity-ai langchain cursor cline model-releases benchmarking long-context model-pricing model-integration voice performance scaling gpu-optimization elonmusk aravsrinivas igor_babuschkin yuchenj_uw
xAI launched Grok 4 and Grok 4 Heavy, large language models rumored to have 2.4 trillion parameters and trained with 100x more compute than Grok 2 on 100k H100 GPUs. Grok 4 achieved new state-of-the-art results on benchmarks like ARC-AGI-2 (15.9%), HLE (50.7%), and Vending-Bench, outperforming models such as Claude 4 Opus. The model supports a 256K context window and is priced at $3.00/M input tokens and $15.00/M output tokens. It is integrated into platforms like Cursor, Cline, LangChain, and Perplexity Pro/Max. The launch was accompanied by a controversial voice mode and sparked industry discussion about xAI's rapid development pace, with endorsements from figures like Elon Musk and Arav Srinivas.
not much happened today
grok-4 smollm3 t5gemma claude-3.7-sonnet deepseek-r1 langchain openai google-deepmind perplexity xai microsoft huggingface anthropic agentic-ai model-controversy open-source model-release alignment fine-tuning long-context multimodality model-research aravsrinivas clementdelangue _akhaliq
LangChain is nearing unicorn status, while OpenAI and Google DeepMind's Gemini 3 Pro models are launching soon. Perplexity rolls out its agentic browser Comet to waitlists, offering multitasking and voice command features. xAI's Grok-4 update sparked controversy due to offensive outputs, drawing comparisons to Microsoft's Tay bot and resulting in regional blocks. Hugging Face released SmolLM3, a 3B parameter open-source model with state-of-the-art reasoning and long context capabilities. Google introduced T5Gemma encoder-decoder models, a significant update in this model category. Anthropic investigates "alignment faking" in language models, focusing on safety concerns with models like Claude 3.7 Sonnet and DeepSeek-R1. "Grok 3 had high reasoning, Grok 4 has heil reasoning" was a notable user comment on the controversy.
Zuck goes Superintelligence Founder Mode: $100M bonuses + $100M+ salaries + NFDG Buyout?
llama-4 maverick scout minimax-m1 afm-4.5b chatgpt midjourney-v1 meta-ai-fair openai deeplearning-ai essential-ai minimax arcee midjourney long-context multimodality model-release foundation-models dataset-release model-training video-generation enterprise-ai model-architecture moe prompt-optimization sama nat dan ashvaswani clementdelangue amit_sangani andrewyng _akhaliq
Meta AI is reportedly offering 8-9 figure signing bonuses and salaries to top AI talent, confirmed by Sam Altman. They are also targeting key figures like Nat and Dan from the AI Grant fund for strategic hires. Essential AI released the massive 24-trillion-token Essential-Web v1.0 dataset with rich metadata and a 12-category taxonomy. DeepLearning.AI and Meta AI launched a course on Llama 4, featuring new MoE models Maverick (400B) and Scout (109B) with context windows up to 10M tokens. MiniMax open-sourced MiniMax-M1, a long-context LLM with a 1M-token window, and introduced the Hailuo 02 video model. OpenAI rolled out "Record mode" for ChatGPT Pro, Enterprise, and Edu on macOS. Arcee launched the AFM-4.5B foundation model for enterprise. Midjourney released its V1 video model enabling image animation. These developments highlight major advances in model scale, long-context reasoning, multimodality, and enterprise AI applications.
Gemini 2.5 Pro/Flash GA, 2.5 Flash-Lite in Preview
gemini-2.5 gemini-2.5-flash-lite gemini-2.5-flash gemini-2.5-pro gemini-2.5-ultra kimi-dev-72b nanonets-ocr-s ii-medical-8b-1706 jan-nano deepseek-r1 minimax-m1 google moonshot-ai deepseek cognitivecompai kling-ai mixture-of-experts multimodality long-horizon-planning benchmarking coding-performance long-context ocr video-generation model-releases tulsee_doshi oriolvinyalsml demishassabis officiallogank _philschmid swyx sainingxie scaling01 gneubig clementdelangue mervenoyann
Gemini 2.5 models are now generally available, including the new Gemini 2.5 Flash-Lite, Flash, Pro, and Ultra variants, featuring sparse Mixture-of-Experts (MoE) transformers with native multimodal support. A detailed 30-page tech report highlights impressive long-horizon planning demonstrated by Gemini Plays Pokemon. The LiveCodeBench-Pro benchmark reveals frontier LLMs struggle with hard coding problems, while Moonshot AI open-sourced Kimi-Dev-72B, achieving state-of-the-art results on SWE-bench Verified. Smaller specialized models like Nanonets-OCR-s, II-Medical-8B-1706, and Jan-nano show competitive performance, emphasizing that bigger models are not always better. DeepSeek-r1 ties for #1 in WebDev Arena, and MiniMax-M1 sets new standards in long-context reasoning. Kling AI demonstrated video generation capabilities.
not much happened today
deepseek-r1-0528 o3 gemini-2.5-pro claude-opus-4 deepseek_ai openai gemini meta-ai-fair anthropic x-ai ollama hugging-face alibaba bytedance xiaomi reasoning reinforcement-learning benchmarking quantization local-inference model-evaluation open-weights transparency post-training agentic-benchmarks long-context hallucination-detection teortaxestex wenfeng danielhanchen awnihannun reach_vb abacaj
DeepSeek R1-0528 release brings major improvements in reasoning, hallucination reduction, JSON output, and function calling, matching or surpassing closed models like OpenAI o3 and Gemini 2.5 Pro on benchmarks such as Artificial Analysis Intelligence Index, LiveBench, and GPQA Diamond. The model ranks #2 globally in open weights intelligence, surpassing Meta AI, Anthropic, and xAI. Open weights and technical transparency have fueled rapid adoption across platforms like Ollama and Hugging Face. Chinese AI labs including DeepSeek, Alibaba, ByteDance, and Xiaomi now match or surpass US labs in model releases and intelligence, driven by open weights strategies. Reinforcement learning post-training is critical for intelligence gains, mirroring trends seen at OpenAI. Optimized quantization techniques (1-bit, 4-bit) and local inference enable efficient experimentation on consumer hardware. New benchmarks like LisanBench test knowledge, planning, memory, and long-context reasoning, with OpenAI o3 and Claude Opus 4 leading. Discussions highlight concerns about benchmark contamination and overemphasis on RL-tuned gains.
not much happened today
deepseek-r1-0528 pali-gemma-2 gemma-3 shieldgemma-2 txgemma gemma-3-qat gemma-3n-preview medgemma dolphingemma signgemma claude-4 opus-4 claude-sonnet-4 codestral-embed bagel qwen nemotron-cortexa gemini-2.5-pro deepseek-ai huggingface gemma claude bytedance qwen nemotron sakana-ai-labs benchmarking model-releases multimodality code-generation model-performance long-context reinforcement-learning model-optimization open-source yuchenj_uw _akhaliq clementdelangue osanseviero alexalbert__ guillaumelample theturingpost lmarena_ai epochairesearch scaling01 nrehiew_ ctnzr
DeepSeek R1 v2 model released with availability on Hugging Face and inference partners. The Gemma model family continues prolific development including PaliGemma 2, Gemma 3, and others. Claude 4 and its variants like Opus 4 and Claude Sonnet 4 show top benchmark performance, including new SOTA on ARC-AGI-2 and WebDev Arena. Codestral Embed introduces a 3072-dimensional code embedder. BAGEL, an open-source multimodal model by ByteDance, supports reading, reasoning, drawing, and editing with long mixed contexts. Benchmarking highlights include Nemotron-CORTEXA topping SWEBench and Gemini 2.5 Pro performing on VideoGameBench. Discussions on random rewards effectiveness focus on Qwen models. "Opus 4 NEW SOTA ON ARC-AGI-2. It's happening - I was right" and "Claude 4 launch has dev moving at a different pace" reflect excitement in the community.
Qwen 3: 0.6B to 235B MoE full+base models that beat R1 and o1
qwen-3 qwen3-235b-a22b qwen3-30b-a3b deepseek-r1 o1 o3-mini grok-3 gemini-2.5-pro alibaba google-deepmind deepseek mistral-ai mixture-of-experts reinforcement-learning benchmarking model-release model-architecture long-context multi-agent-systems inference dataset-release awnihannun prince_canuma actuallyisaak oriolvinyalsml iscienceluvr reach_vb teortaxestex omarsar0
Qwen 3 has been released by Alibaba featuring a range of models including two MoE variants, Qwen3-235B-A22B and Qwen3-30B-A3B, which demonstrate competitive performance against top models like DeepSeek-R1, o1, o3-mini, Grok-3, and Gemini-2.5-Pro. The models introduce an "enable_thinking=True" mode with advanced soft switching for inference scaling. The release is notable for its Apache 2.0 license and broad inference platform support including MCP. The dataset improvements and multi-stage RL post-training contribute to performance gains. Meanwhile, Gemini 2.5 Pro from Google DeepMind shows strong coding and long-context reasoning capabilities, and DeepSeek R2 is anticipated soon. Twitter discussions highlight Qwen3's finegrained MoE architecture, large context window, and multi-agent system applications.
gpt-image-1 - ChatGPT's imagegen model, confusingly NOT 4o, now available in API
gpt-image-1 o3 o4-mini gpt-4.1 eagle-2.5-8b gpt-4o qwen2.5-vl-72b openai nvidia hugging-face x-ai image-generation content-moderation benchmarking long-context multimodality model-performance supercomputing virology video-understanding model-releases kevinweil lmarena_ai _philschmid willdepue arankomatsuzaki epochairesearch danhendrycks reach_vb mervenoyann _akhaliq
OpenAI officially launched the gpt-image-1 API for image generation and editing, supporting features like alpha channel transparency and a "low" content moderation policy. OpenAI's models o3 and o4-mini are leading in benchmarks for style control, math, coding, and hard prompts, with o3 ranking #1 in several categories. A new benchmark called Vending-Bench reveals performance variance in LLMs on extended tasks. GPT-4.1 ranks in the top 5 for hard prompts and math. Nvidia's Eagle 2.5-8B matches GPT-4o and Qwen2.5-VL-72B in long-video understanding. AI supercomputer performance doubles every 9 months, with xAI's Colossus costing an estimated $7 billion and the US dominating 75% of global performance. The Virology Capabilities Test shows OpenAI's o3 outperforms 94% of expert virologists. Nvidia also released the Describe Anything Model (DAM), a multimodal LLM for detailed image and video captioning, now available on Hugging Face.
not much happened today
nemotron-h nvidia-eagle-2.5 gpt-4o qwen2.5-vl-72b gemini-2.5-flash gemini-2.0-pro gemini-exp-1206 gemma-3 qwen2.5-32b deepseek-r1-zero-32b uni3c seedream-3.0 adobe-dragon kimina-prover qwen2.5-72b bitnet-b1.58-2b4t nvidia deepseek hugging-face alibaba bytedance adobe transformers model-optimization multimodality long-context reinforcement-learning torch-compile image-generation diffusion-models distributional-rewards model-efficiency model-training native-quantization sampling-techniques philschmid arankomatsuzaki osanseviero iScienceLuvr akhaliq
Nemotron-H model family introduces hybrid Mamba-Transformer models with up to 3x faster inference and variants including 8B, 56B, and a compressed 47B model. Nvidia Eagle 2.5 is a frontier VLM for long-context multimodal learning, matching GPT-4o and Qwen2.5-VL-72B on long-video understanding. Gemini 2.5 Flash shows improved dynamic thinking and cost-performance, outperforming previous Gemini versions. Gemma 3 now supports torch.compile for about 60% faster inference on consumer GPUs. SRPO using Qwen2.5-32B surpasses DeepSeek-R1-Zero-32B on benchmarks with reinforcement learning only. Alibaba's Uni3C unifies 3D-enhanced camera and human motion controls for video generation. Seedream 3.0 by ByteDance is a bilingual image generation model with high-resolution outputs up to 2K. Adobe DRAGON optimizes diffusion generative models with distributional rewards. Kimina-Prover Preview is an LLM trained with reinforcement learning from Qwen2.5-72B, achieving 80.7% pass@8192 on miniF2F. BitNet b1.58 2B4T is a native 1-bit LLM with 2B parameters trained on 4 trillion tokens, matching full-precision LLM performance with better efficiency. Antidistillation sampling counters unwanted model distillation by modifying reasoning traces from frontier models.
GPT 4.1: The New OpenAI Workhorse
gpt-4.1 gpt-4.1-mini gpt-4.1-nano gpt-4o gemini-2.5-pro openai llama-index perplexity-ai google-deepmind coding instruction-following long-context benchmarks model-pricing model-integration model-deprecation sama kevinweil omarsar0 aidan_mclau danhendrycks polynoamial scaling01 aravsrinivas lmarena_ai
OpenAI released GPT-4.1, including GPT-4.1 mini and GPT-4.1 nano, highlighting improvements in coding, instruction following, and handling long contexts up to 1 million tokens. The model achieves a 54 score on SWE-bench verified and shows a 60% improvement over GPT-4o on internal benchmarks. Pricing for GPT-4.1 nano is notably low at $0.10/1M input and $0.40/1M output. GPT-4.5 Preview is being deprecated in favor of GPT-4.1. Integration support includes Llama Index with day 0 support. Some negative feedback was noted for GPT-4.1 nano. Additionally, Perplexity's Sonar API ties with Gemini-2.5 Pro for the top spot in the LM Search Arena leaderboard. New benchmarks like MRCR and GraphWalks were introduced alongside updated prompting guides and cookbooks.
not much happened today
gpt-4o deepseek-v3 claude-3.7-sonnet o3-mini gemini-2.5-pro openai deepseek anthropic google-deepmind togethercompute hypertecgroup coreweave cursor-ai windsurf-ai coding instruction-following image-generation policy-compliance long-context audio-processing video-processing gpu-clusters ai-infrastructure api-access sama kevinweil joannejang nrehiew_ giffmana _philschmid scaling01 saranormous
GPT-4o was praised for its improved coding, instruction following, and freedom, becoming the leading non-reasoning coding model surpassing DeepSeek V3 and Claude 3.7 Sonnet in coding benchmarks, though it still lags behind reasoning models like o3-mini. Concerns about policy compliance in image generation were noted, with efforts to improve adherence. Gemini 2.5 Pro was highlighted for its advanced audio and video understanding, long context capabilities, and integration with platforms like Cursor AI and Windsurf AI. AI infrastructure developments include a partnership between Together AI and Hypertec Group to deliver large-scale GPU clusters, and CoreWeave's IPO was celebrated for advancing AI infrastructure. GPU and TPU usage is expected to increase significantly. "GPT-4o's transparency and background generation feature" and "Gemini 2.5 Pro scored above 50% on Simple-Bench AI Explanation" were key highlights.
LLaDA: Large Language Diffusion Models
llada-8b llama-3-8b step-video-t2v-30b step-audio-chat-132b llama-2-7b stepfun-ai scale-ai cambridge llamaindex diffusion-models text-generation multimodality video-generation voice-processing benchmarking instruction-following model-scaling gpu-usage long-context multi-turn-dialogue arankomatsuzaki _akhaliq omarsar0 iscienceluvr gallabytes maximelabonne reach_vb
LLaDA (Large Language Diffusion Model) 8B is a breakthrough diffusion-based language model that rivals LLaMA 3 8B while training on 7x fewer tokens (2 trillion tokens) and using 0.13 million H800 GPU hours. It introduces a novel text generation approach by predicting uniformly masked tokens in a diffusion process, enabling multi-turn dialogue and instruction-following. Alongside, StepFun AI released two major models: Step-Video-T2V 30B, a text-to-video model generating up to 204 frames with high coherence and motion quality, and Step-Audio-Chat 132B, a voice-to-voice model. Additionally, challenging multimodal benchmarks like Scale AI's EnigmaEval and Cambridge's ZeroBench highlight current frontier models scoring zero, emphasizing the difficulty of these tasks. The community also noted the return of diffusion models in language modeling, a previously speculative architecture now scaled successfully.
Project Stargate: $500b datacenter (1.7% of US GDP) and Gemini 2 Flash Thinking 2
gemini-2.0-flash deepseek-r1 qwen-32b openai softbank oracle arm microsoft nvidia huggingface deepseek-ai long-context quantization code-interpretation model-distillation open-source agi-research model-performance memory-optimization noam-shazeer liang-wenfeng
Project Stargate, a US "AI Manhattan project" led by OpenAI and Softbank, supported by Oracle, Arm, Microsoft, and NVIDIA, was announced with a scale comparable to the original Manhattan project costing $35B inflation adjusted. Despite Microsoft's reduced role as exclusive compute partner, the project is serious but not immediately practical. Meanwhile, Noam Shazeer revealed a second major update to Gemini 2.0 Flash Thinking, enabling 1M token long context usable immediately. Additionally, AI Studio introduced a new code interpreter feature. On Reddit, DeepSeek R1, a distillation of Qwen 32B, was released for free on HuggingChat, sparking discussions on self-hosting, performance issues, and quantization techniques. DeepSeek's CEO Liang Wenfeng highlighted their focus on fundamental AGI research, efficient MLA architecture, and commitment to open-source development despite export restrictions, positioning DeepSeek as a potential alternative to closed-source AI trends.
Titans: Learning to Memorize at Test Time
minimax-01 gpt-4o claude-3.5-sonnet internlm3-8b-instruct transformer2 google meta-ai-fair openai anthropic langchain long-context mixture-of-experts self-adaptive-models prompt-injection agent-authentication diffusion-models zero-trust-architecture continuous-adaptation vision agentic-systems omarsar0 hwchase17 abacaj hardmaru rez0__ bindureddy akhaliq saranormous
Google released a new paper on "Neural Memory" integrating persistent memory directly into transformer architectures at test time, showing promising long-context utilization. MiniMax-01 by @omarsar0 features a 4 million token context window with 456B parameters and 32 experts, outperforming GPT-4o and Claude-3.5-Sonnet. InternLM3-8B-Instruct is an open-source model trained on 4 trillion tokens with state-of-the-art results. Transformerยฒ introduces self-adaptive LLMs that dynamically adjust weights for continuous adaptation. Advances in AI security highlight the need for agent authentication, prompt injection defenses, and zero-trust architectures. Tools like Micro Diffusion enable budget-friendly diffusion model training, while LeagueGraph and Agent Recipes support open-source social media agents.
ModernBert: small new Retriever/Classifier workhorse, 8k context, 2T tokens,
modernbert gemini-2.0-flash-thinking o1 llama answerdotai lightonio hugging-face google-deepmind openai meta-ai-fair figure encoder-only-models long-context alternating-attention natural-language-understanding reasoning robotics-simulation physics-engine humanoid-robots model-performance model-releases jeremyphoward alec-radford philschmid drjimfan bindureddy
Answer.ai/LightOn released ModernBERT, an updated encoder-only model with 8k token context, trained on 2 trillion tokens including code, with 139M/395M parameters and state-of-the-art performance on retrieval, NLU, and code tasks. It features Alternating Attention layers mixing global and local attention. Gemini 2.0 Flash Thinking debuted as #1 in Chatbot Arena, and the O1 model scored top in reasoning benchmarks. Llama downloads surpassed 650 million, doubling in 3 months. OpenAI launched desktop app integrations with voice capabilities. Figure delivered its first humanoid robots commercially. Advances in robotics simulation and a new physics engine Genesis claiming 430,000x faster than real-time were highlighted.
not much happened today
llama mistral openai decagon sierra togethercompute vertical-saas funding protein-structure-prediction lora self-supervised-learning model-optimization neural-architecture-search model-evaluation ethics transformers multi-agent-systems long-context mira-murati demis-hassabis clement-delangue john-o-whitaker yann-lecun francois-chollet ajeya-cotra rohan-paul adcock-brett
Vertical SaaS agents are gaining rapid consensus as the future of AI applications, highlighted by Decagon's $100m funding and Sierra's $4b round. OpenAI alumni are actively raising venture capital and forming new startups, intensifying competition in the AI market. Demis Hassabis celebrated the Nobel Prize recognition for AlphaFold2, a breakthrough in protein structure prediction. Advances in AI models include techniques like LoRA projectors and annealing on high-quality data, while discussions emphasize the need for high-bandwidth sensory inputs beyond language for common sense learning. New methods like LoLCATs aim to optimize transformer models such as Llama and Mistral for efficiency. Ethical concerns about AI agents performing harmful tasks remain under investigation. The AI community continues to explore model evaluation challenges and optimization frameworks like LPZero for neural architecture search.
Not much (in AI) happened this weekend
llama-3.1-8b llama-3.2 chatgpt movie-gen openai meta-ai-fair google-deepmind microsoft x-ai spacex harvard nvidia long-context feature-prediction-loss ai-agents privacy text-to-video text-to-image humanoid-robots gpu-deployment media-foundation-models ai-research-labs sam-altman yann-lecun rasbt bindureddy andrej-karpathy soumithchintala svpino adcock_brett rohanpaul_ai
OpenAI introduced an "edit this area" feature for image generation, praised by Sam Altman. Yann LeCun highlighted a NYU paper improving pixel generation with feature prediction loss using pre-trained visual encoders like DINOv2. Long-context LLMs such as llama-3.1-8b and llama-3.2 variants now support up to 131k tokens, offering alternatives to RAG systems. Bindu Reddy announced AI agents capable of building and deploying code from English instructions, signaling AI's replacement of SQL and potential impact on Python. SpaceX's successful Starship rocket catch was celebrated by Andrej Karpathy and others, with Soumith Chintala praising SpaceX's efficient, low-bureaucracy research approach. Privacy concerns arose from Harvard students' AI glasses, I-XRAY, which can reveal personal information. Meta AI FAIR's Movie Gen model advances media foundation models with high-quality text-to-image and video generation, including synced audio. Humanoid robots like Ameca and Azi now engage in expressive conversations using ChatGPT. xAI rapidly deployed 100K Nvidia H100 GPUs in 19 days, with CEO Jensen Huang commending Elon Musk. Leading AI research labs compared include Meta-FAIR, Google DeepMind, and Microsoft Research. Skepticism about LLM intelligence was voiced by Sam Pino, emphasizing limitations in novel problem-solving despite strong memorization.
not much happened today
aria o1-preview o1-mini gemini-1.5-pro gemini-1.5-flash gemini-1.5 claude-3.5-sonnet rhymes-ai openai anthropic google meta-ai-fair oxylabs multimodality mixture-of-experts long-context retrieval-augmented-generation benchmarking software-engineering llm-evaluation prompt-engineering web-scraping python production-applications mervenoyann osanseviero dbrxmosaicai ylecun ofirpress clefourrier omarsar0 rohanpaul_ai svpino finbarrtimbers _philschmid
Rhymes AI released Aria, a new 25.3B parameter multimodal MoE model supporting text, code, image, and video with a 64k token context window and Apache-2.0 license. OpenAI's o1-preview and o1-mini models show consistent improvement over Anthropic and Google Gemini 1.5 Pro/Flash on long context RAG benchmarks up to 128k tokens, while Google Gemini 1.5 models excel at extreme context lengths up to 2 million tokens. Meta AI expanded rollout to 21 countries with new language support but remains unavailable in the EU. The one-year anniversary of SWE-bench benchmark for software engineering tasks was celebrated, alongside the introduction of SWE-bench Multimodal. New AI tools include OxyCopilot by Oxylabs for web scraping, Taipy for Python-based production apps, and Latitude for prompt engineering. Industry insights highlight changing AI funding dynamics and OpenAI's strategic focus on consumer products like ChatGPT. "all recaps done by Claude 3.5 Sonnet, best of 4 runs."
a calm before the storm
o1 o1-mini qwen2.5 gpt-4 llama-2-70b llama-7b anthropic openai alibaba microsoft blackrock groq aramco disney eth-zurich pudu-robotics slack long-context kv-cache-quantization diffusion-models reinforcement-learning robotics ai-integration multilinguality model-benchmarking model-performance model-optimization adcock_brett philschmid rohanpaul_ai jvnixon kateclarktweets sama
Anthropic is raising funds at a valuation up to $40 billion ahead of anticipated major releases. OpenAI launched new reasoning models o1 and o1-mini, with increased rate limits and a multilingual MMLU benchmark. Alibaba released the open-source Qwen2.5 model supporting 29+ languages, showing competitive performance to gpt-4 at lower cost. Microsoft and Blackrock plan to invest $30 billion in AI data centers, with Groq partnering with Aramco to build the world's largest AI inference center. Robotics advances include Disney Research and ETH Zurich's diffusion-based motion generation for robots and Pudu Robotics' semi-humanoid robot. Slack and Microsoft introduced AI-powered agents integrated into their platforms. Research highlights include long-context scaling for llama-2-70b using Dual Chunk Attention and KV cache quantization enabling 1 million token context on llama-7b models.
not much happened today
llama-3 o1 deepseek-2.5 gpt-4 claude-3.5-sonnet 3dtopia-xl cogvideox anthropic meta-ai-fair openai deepseek-ai llamaindex langchainai retrieval-augmented-generation prompt-caching multimodality multi-agent-systems reasoning diffusion-models image-to-video prompting enterprise-ai agentic-ai long-context model-evaluation caching model-cost-efficiency
Anthropic introduced a RAG technique called Contextual Retrieval that reduces retrieval failure rates by 67% using prompt caching. Meta is teasing multimodal Llama 3 ahead of Meta Connect. OpenAI is hiring for a multi-agent research team focusing on improved AI reasoning with their o1 models, which have sparked mixed reactions. DeepSeek 2.5 is noted as a cost-effective alternative to GPT-4 and Claude 3.5 sonnet. New models like 3DTopia-XL for 3D asset generation and CogVideoX for image-to-video conversion were highlighted. Techniques to boost reasoning by re-reading questions and combining retrieval with prompt caching were shared. Industry insights emphasize the necessity of AI adoption in enterprises and the disruption of traditional ML businesses. Tools like LangChainAI's LangGraph Templates and LlamaIndex's LlamaParse Premium enhance agentic applications and multimodal content extraction. Discussions on LLM evals and caching highlight production challenges and improvements. "Companies not allowing developers to use AI are unlikely to succeed" was a key sentiment.
Everybody shipped small things this holiday weekend
gpt-4o-voice gemini claude jamba-1.5 mistral-nemo-minitron-8b xai google anthropic openai cognition ai21-labs nvidia langchain fine-tuning long-context parameter-efficient-fine-tuning latex-rendering real-time-audio virtual-try-on resource-tags low-code ai-agents workspace-organization model-benchmarking dario-amodei scott-wu fchollet svpino
xAI announced the Colossus 100k H100 cluster capable of training an FP8 GPT-4 class model in 4 days. Google introduced Structured Output for Gemini. Anthropic discussed Claude's performance issues possibly due to API prompt modifications. OpenAI enhanced controls for File Search in their Assistants API. Cognition and Anthropic leaders appeared on podcasts. The viral Kwai-Kolors virtual try-on model and the open-source real-time audio conversational model Mini-Omni (similar to gpt-4o-voice) were released. Tutorials on parameter-efficient fine-tuning with LoRA and QLoRA, long-context embedding challenges, and Claude's LaTeX rendering feature were highlighted. AI21 Labs released Jamba 1.5 models with a 256K context window and faster long-context performance. NVIDIA debuted Mistral-Nemo-Minitron-8B on the Open LLM Leaderboard. LangChain introduced resource tags for workspace organization, and a low-code AI app toolkit was shared by svpino. Legal AI agents and financial agent evaluations using LangSmith were also featured.
not much happened today
llama-3-1 claude-3-5-sonnet llama-3-1-405b ltm-2-mini qwen2-vl gpt-4o-mini meta-ai-fair hugging-face magic-ai-labs lmsys alibaba openai long-context style-control multimodality ai-safety model-evaluation web-crawling pdf-processing ai-hype-cycles call-center-automation sam-altman ajeya-cotra fchollet rohanpaul_ai philschmid
Meta announced significant adoption of LLaMA 3.1 with nearly 350 million downloads on Hugging Face. Magic AI Labs introduced LTM-2-Mini, a long context model with a 100 million token context window, and a new evaluation method called HashHop. LMSys added style control to their Chatbot Arena leaderboard, improving rankings for models like Claude 3.5 Sonnet and LLaMA 3.1 405B. Alibaba released Qwen2-VL, a multimodal LLM under Apache 2.0 license, competitive with GPT-4o mini. OpenAI CEO Sam Altman announced collaboration with the US AI Safety Institute for pre-release model testing. Discussions on AI safety and potential AI takeover risks were highlighted by Ajeya Cotra. Tools like firecrawl for web crawling and challenges in PDF processing were noted. AI hype cycles and market trends were discussed by Franรงois Chollet, and potential AI disruption in call centers was shared by Rohan Paul.
Summer of Code AI: $1.6b raised, 1 usable product
ltm-2 llama-3-1-405b gemini-advanced cognition poolside codeium magic google-deepmind nvidia google-cloud long-context model-efficiency custom-hardware cuda training-stack gpu-scaling neural-world-models diffusion-models quantization nat-friedman ben-chess rohan-paul
Code + AI is emphasized as a key modality in AI engineering, highlighting productivity and verifiability benefits. Recent major funding rounds include Cognition AI raising $175M, Poolside raising $400M, Codeium AI raising $150M, and Magic raising $320M. Magic announced their LTM-2 model with a 100 million token context window, boasting efficiency improvements over Llama 3.1 405B by about 1000x cheaper in sequence-dimension algorithm and drastically lower memory requirements. Magic's stack is built from scratch with custom CUDA and no open-source foundations, partnered with Google Cloud and powered by NVIDIA H100 and GB200 GPUs, aiming to scale to tens of thousands of GPUs. Google DeepMind revealed updates to Gemini Advanced with customizable expert "Gems." Neural Game Engines like GameNGen can run DOOM in a diffusion model trained on 0.9B frames. The content also references LLM quantization research by Rohan Paul.
CogVideoX: Zhipu's Open Source Sora
cogvideox llama-3-1 llama-3-405b moondream phi-3.5 llama-rank zhipu-ai alibaba meta-ai-fair google hugging-face nvidia togethercompute salesforce video-generation serverless-computing vision document-vqa text-vqa mixture-of-experts retrieval-augmented-generation long-context model-routing webgpu background-removal long-form-generation superposition-prompting rohanpaul_ai philschmid vikhyatk algo_diver jayalammar davidsholz
Zhipu AI, Alibaba's AI arm and China's 3rd largest AI lab, released the open 5B video generation model CogVIdeoX, which can run without GPUs via their ChatGLM web and desktop apps. Meta AI announced trust & safety research and CyberSecEval 3 alongside the release of Llama 3.1, with Llama 3 405B now available serverless on Google Cloud Vertex AI and Hugging Face x NVIDIA NIM API. Updates include Moondream, an open vision-language model improving DocVQA and TextVQA tasks, and the lightweight MoE chat model Phi-3.5 with 16x3.8B parameters. Together Compute introduced the Rerank API featuring Salesforce's LlamaRank model for document and code ranking. Research highlights include superposition prompting for RAG without fine-tuning, the AgentWrite pipeline for long-form content generation over 20,000 words, and a comparison showing Long Context methods outperform RAG at higher costs. Tools include Not Diamond, an AI model router, AI command line interfaces, and an open-source WebGPU background removal tool. "You don't even need GPUs to run it," referring to CogVIdeoX.
not much happened this weekend
jamba-1.5 dream-machine-1.5 ideogram-v2 mistral-nemo-minitron-8b mistral-7b llama-3-8b nous-research cursor-ai gdm george-hotz agibot unitree eth-zurich disney uc-san-diego ai21-labs luma-labs ideogram nvidia mistral-ai meta-ai-fair distributed-ai optimizer inter-gpu-communication low-latency-training open-source humanoid-robots robotics physics-based-motion teleoperation multilingual-models long-context text-to-video text-to-image model-performance george-hotz adcock_brett aman
Nous Research announced DisTrO, a new optimizer that drastically reduces inter-GPU communication by 1000x to 10,000x enabling efficient training on slow networks, offering an alternative to GDM's DiLoCo. Cursor AI gained viral attention from an 8-year-old user and announced a new fundraise, with co-host Aman returning to their podcast. George Hotz launched tinybox for sale. In robotics, AGIBOT revealed 5 new humanoid robots with open-source plans, and Unitree showcased its G1 humanoid robot nearing mass production at $16,000. ETH Zurich and Disney developed an AI system for physics-based robot motion generation from text or images. UC San Diego released ACE, an open-source teleoperation system for controlling multiple robots. AI21 Labs unveiled Jamba 1.5, a multilingual model with 256k context length and permissive licensing. Luma Labs released Dream Machine 1.5 for improved text-to-video generation. Ideogram launched v2 of its text-to-image model with near-perfect text generation. Nvidia and Mistral released Mistral-NeMo-Minitron 8B, a small model outperforming Mistral-7B and llama-3-8b on the Open LLM leaderboard.
Nvidia Minitron: LLM Pruning and Distillation updated for Llama 3.1
llama-3-1-8b llama-3-1 jamba-1.5 claude-3 dracarys-70b dracarys-72b mistral-nemo-minitron-8b mistral-7b nvidia meta-ai-fair ai21-labs anthropic hugging-face pruning knowledge-distillation weight-pruning activation-based-pruning width-pruning kl-divergence teacher-correction prompt-optimization multilinguality long-context mixture-of-experts model-fine-tuning
Nvidia and Meta researchers updated their Llama 3 results with a paper demonstrating the effectiveness of combining weight pruning and knowledge distillation to reduce training costs by training only the largest model from scratch and deriving smaller models via pruning and distillation. The process involves teacher correction, activation-based pruning (favoring width pruning), and retraining with distillation using KL Divergence loss, resulting in better-performing models at comparable sizes. However, distillation incurs some accuracy tradeoffs. Additionally, AI21 Labs launched Jamba 1.5, a hybrid SSM-Transformer MoE model with large context windows and multilingual support. Anthropic updated Claude 3 with LaTeX rendering and prompt caching. An open-source coding-focused LLM, Dracarys, was released in 70B and 72B sizes, showing improved coding performance. The Mistral Nemo Minitron 8B model outperforms Llama 3.1 8B and Mistral 7B on the Hugging Face leaderboard, highlighting pruning and distillation benefits. Research on prompt optimization reveals the complexity of prompt search spaces and the surprising effectiveness of simple algorithms like AutoPrompt/GCG.
super quiet day
jamba-1.5 phi-3.5 dracarys llama-3-1-70b llama-3-1 ai21-labs anthropic stanford hugging-face langchain qdrant aws elastic state-space-models long-context benchmarking ai-safety virtual-environments multi-agent-systems resource-management community-engagement model-performance bindu-reddy rohanpaul_ai jackclarksf danhendrycks reach_vb iqdotgraph
AI21 Labs released Jamba 1.5, a scaled-up State Space Model optimized for long context windows with 94B parameters and up to 2.5X faster inference, outperforming models like Llama 3.1 70B on benchmarks. The Phi-3.5 model was praised for its safety and performance, while Dracarys, a new 70B open-source coding model announced by Bindu Reddy, claims superior benchmarks over Llama 3.1 70B. Discussions on California's SB 1047 AI safety legislation involve Stanford and Anthropic, highlighting a balance between precaution and industry growth. Innovations include uv virtual environments for rapid setup, LangChain's LangSmith resource tags for project management, and multi-agent systems in Qdrant enhancing data workflows. Community events like the RAG workshop by AWS, LangChain, and Elastic continue to support AI learning and collaboration. Memes remain a popular way to engage with AI industry culture.
Gemini Live
gemini-1.5-pro genie falcon-mamba gemini-1.5 llamaindex google anthropic tii supabase perplexity-ai llamaindex openai hugging-face multimodality benchmarking long-context retrieval-augmented-generation open-source model-releases model-integration model-performance software-engineering linear-algebra hugging-face-hub debugging omarsar0 osanseviero dbrxmosaicai alphasignalai perplexity_ai _jasonwei svpino
Google launched Gemini Live on Android for Gemini Advanced subscribers during the Pixel 9 event, featuring integrations with Google Workspace apps and other Google services. The rollout began on 8/12/2024, with iOS support planned. Anthropic released Genie, an AI software engineering system achieving a 57% improvement on SWE-Bench. TII introduced Falcon Mamba, a 7B attention-free open-access model scalable to long sequences. Benchmarking showed that longer context lengths do not always improve Retrieval-Augmented Generation. Supabase launched an AI-powered Postgres service dubbed the "ChatGPT of databases," fully open source. Perplexity AI partnered with Polymarket to integrate real-time probability predictions into search results. A tutorial demonstrated a multimodal recipe recommender using Qdrant, LlamaIndex, and Gemini. An OpenAI engineer shared success tips emphasizing debugging and hard work. The connection between matrices and graphs in linear algebra was highlighted for insights into nonnegative matrices and strongly connected components. Keras 3.5.0 was released with Hugging Face Hub integration for model saving and loading.
Llama 3.1: The Synthetic Data Model
llama-3-405b llama-3-1 llama-3 meta-ai-fair groq fireworks synthetic-data fine-tuning reinforcement-learning multilinguality long-context tool-use code-generation math model-licensing inference-speed model-deployment bindureddy thomas
Meta AI has released Llama 3.1, including a 405B parameter model that triggers regulatory considerations like the EU AI Act and SB 1047. The model incorporates extensive synthetic data techniques for code, math, multilinguality, long context, and tool use fine-tuning, with RLHF using synthetic preference data from Llama 2. The launch was coordinated across major inference providers, with Groq demonstrating 750 tokens per second inference speed and Fireworks leading in pricing. The updated license explicitly allows synthetic data generation, marking a significant step in open frontier-class LLMs and cost-efficiency improvements since March.
Mini, Nemo, Turbo, Lite - Smol models go brrr (GPT4o-mini version)
gpt-4o-mini deepseek-v2-0628 mistral-nemo llama-8b openai deepseek-ai mistral-ai nvidia meta-ai-fair hugging-face langchain keras cost-efficiency context-windows open-source benchmarking neural-networks model-optimization text-generation fine-tuning developer-tools gpu-support parallelization cuda-integration multilinguality long-context article-generation liang-wenfeng
OpenAI launched the GPT-4o Mini, a cost-efficient small model priced at $0.15 per million input tokens and $0.60 per million output tokens, aiming to replace GPT-3.5 Turbo with enhanced intelligence but some performance limitations. DeepSeek open-sourced DeepSeek-V2-0628, topping the LMSYS Chatbot Arena Leaderboard and emphasizing their commitment to contributing to the AI ecosystem. Mistral AI and NVIDIA released the Mistral NeMo, a 12B parameter multilingual model with a record 128k token context window under an Apache 2.0 license, sparking debates on benchmarking accuracy against models like Meta Llama 8B. Research breakthroughs include the TextGrad framework for optimizing compound AI systems via textual feedback differentiation and the STORM system improving article writing by 25% through simulating diverse perspectives and addressing source bias. Developer tooling trends highlight LangChain's evolving context-aware reasoning applications and the Modular ecosystem's new official GPU support, including discussions on Mojo and Keras 3.0 integration.
Not much happened today.
phi-3-mini gpt4all-3.0 yi-large meta-3d-gen meta perplexity-ai microsoft gpt4all langchainai qdrant-engine 3d-generation long-context instruction-following reinforcement-learning-from-human-feedback persona-driven-data-synthesis meta-tuning model-steering memory-retrieval multivector-search universal-query-api rohanpaul_ai andriy_mulyar cwolferesearch sarahookr
Meta introduced Meta 3D Gen, a system for end-to-end generation of 3D assets from text in under 1 minute, producing high-quality 3D assets with detailed textures. Perplexity AI updated Pro Search to handle deeper research with multi-step reasoning and code execution. Microsoft improved Phi-3 Mini with better long-context understanding and instruction following. GPT4All 3.0 launched with support for thousands of models and major OS compatibility, featuring local file chat. Yi-Large model launched on Fireworks AI Playground. Research highlights include the evolution of reinforcement learning from human feedback (RLHF), persona-driven data synthesis using a billion diverse personas, meta-tuning for few-shot generalization, and steering vectors for model behavior control. Tools updates include LangSmith improving memory retrieval and Qdrant Engine v1.10 adding universal query API and multivector search.
Nemotron-4-340B: NVIDIA's new large open models, built on syndata, great for syndata
nemotron-4-340b mixtral llama-3 gemini-1.5 gpt-4o mamba-2-hybrid-8b samba-3.8b-instruct dolphin-2.9.3 faro-yi-9b-dpo nvidia hugging-face mistral-ai llamaindex cohere gemini mistral synthetic-data model-alignment reward-models fine-tuning long-context model-scaling inference-speed mixture-of-agents open-source-models model-training instruction-following context-windows philipp-schmid bryan-catanzaro oleksii-kuchaiev rohanpaul_ai cognitivecompai _philschmid 01ai_yi
NVIDIA has scaled up its Nemotron-4 model from 15B to a massive 340B dense model, trained on 9T tokens, achieving performance comparable to GPT-4. The model alignment process uses over 98% synthetic data, with only about 20K human-annotated samples for fine-tuning and reward model training. The synthetic data generation pipeline is open-sourced, including synthetic prompts and preference data generation. The base and instruct versions outperform Mixtral and Llama 3, while the reward model ranks better than Gemini 1.5, Cohere, and GPT-4o. Other notable models include Mamba-2-Hybrid 8B, which is up to 8x faster than Transformers and excels on long-context tasks, Samba-3.8B-instruct for infinite context length with linear complexity, Dolphin-2.9.3 tiny models optimized for low-resource devices, and Faro Yi 9B DPO with a 200K context window running efficiently on 16GB VRAM. The Mixture-of-Agents technique boosts open-source LLMs beyond GPT-4 Omni on AlpacaEval 2.0.
5 small news items
llama-3 xLSTM openai cohere deepmind hugging-face nvidia mistral-ai uncertainty-quantification parameter-efficient-fine-tuning automated-alignment model-efficiency long-context agentic-ai fine-tuning inference-optimization leopold-aschenbrenner will-brown rohanpaul_ai richardmcngo omarsar0 hwchase17 clementdelangue sophiamyang
OpenAI announces that ChatGPT's voice mode is "coming soon." Leopold Aschenbrenner launched a 5-part AGI timelines series predicting a trillion dollar cluster from current AI progress. Will Brown released a comprehensive GenAI Handbook. Cohere completed a $450 million funding round at a $5 billion valuation. DeepMind research on uncertainty quantification in LLMs and an xLSTM model outperforming transformers were highlighted. Studies on the geometry of concepts in LLMs and methods to eliminate matrix multiplication for efficiency gains were shared. Discussions on parameter-efficient fine-tuning (PEFT) and automated alignment of LLMs were noted. New tools include LangGraph for AI agents, LlamaIndex with longer context windows, and Hugging Face's integration with NVIDIA NIM for Llama3. Mistral AI released a fine-tuning API for their models.
1 TRILLION token context, real time, on device?
gemini-1.5-pro gemini-1.5 cartesia mistral-ai scale-ai state-space-models voice-models multimodality model-performance on-device-ai long-context evaluation-leaderboards learning-rate-optimization scientific-publishing research-vs-engineering yann-lecun elon-musk
Cartesia, a startup specializing in state space models (SSMs), launched a low latency voice model outperforming transformer-based models with 20% lower perplexity, 2x lower word error, and 1 point higher NISQA quality. This breakthrough highlights the potential for models that can continuously process and reason over massive streams of multimodal data (text, audio, video) with a trillion token context window on-device. The news also covers recent AI developments including Mistral's Codestral weights release, Schedule Free optimizers paper release, and Scale AI's new elo-style eval leaderboards. Additionally, a debate between yann-lecun and elon-musk on the importance of publishing AI research versus engineering achievements was noted. The Gemini 1.5 Pro/Advanced models were mentioned for their strong performance.
Skyfall
gemini-1.5-pro gemini-1.5-flash yi-1.5 kosmos-2.5 paligemma falcon-2 deepseek-v2 hunyuan-dit gemini-1.5 gemini-1.5-flash yi-1.5 google-deepmind yi-ai microsoft hugging-face langchain maven multimodality mixture-of-experts transformer model-optimization long-context model-performance model-inference fine-tuning local-ai scaling-laws causal-models hallucination-detection model-distillation model-efficiency hamel-husain dan-becker clement-delangue philschmid osanseviero arankomatsuzaki jason-wei rohanpaul_ai
Between 5/17 and 5/20/2024, key AI updates include Google DeepMind's Gemini 1.5 Pro and Flash models, featuring sparse multimodal MoE architecture with up to 10M context and a dense Transformer decoder that is 3x faster and 10x cheaper. Yi AI released Yi-1.5 models with extended context windows of 32K and 16K tokens. Other notable releases include Kosmos 2.5 (Microsoft), PaliGemma (Google), Falcon 2, DeepSeek v2 lite, and HunyuanDiT diffusion model. Research highlights feature an Observational Scaling Laws paper predicting model performance across families, a Layer-Condensed KV Cache technique boosting inference throughput by up to 26ร, and the SUPRA method converting LLMs into RNNs for reduced compute costs. Hugging Face expanded local AI capabilities enabling on-device AI without cloud dependency. LangChain updated its v0.2 release with improved documentation. The community also welcomed a new LLM Finetuning Discord by Hamel Husain and Dan Becker for Maven course users. "Hugging Face is profitable, or close to profitable," enabling $10 million in free shared GPUs for developers.
Not much happened today
gpt-4o gemini-1.5-pro gemini-1.5-flash imagen-3 veo reka-core qwen-1.5-110b openai google-deepmind anthropic rekailabs alibaba salesforce multimodality long-context model-releases reinforcement-learning model-benchmarking text-to-image video-generation ai-assistants ilya-sutskever jakub-pachocki mike-krieger sama
Ilya Sutskever steps down as Chief Scientist at OpenAI after nearly a decade, with Jakub Pachocki named as his successor. Google DeepMind announces Gemini 1.5 Pro and Gemini 1.5 Flash models featuring 2 million token context and improved multimodal capabilities, alongside demos of Project Astra AI assistant, Imagen 3 text-to-image model, and Veo generative video model. GPT-4o tops the VHELM leaderboard and outperforms competitors on LMSYS Chatbot Arena. Reka Core multimodal model with 128K context and Alibaba's Qwen1.5-110B open-source model are released. Salesforce shares an online RLHF recipe.
Evals: The Next Generation
gpt-4 gpt-5 gpt-3.5 phi-3 mistral-7b llama-3 scale-ai mistral-ai reka-ai openai moderna sanctuary-ai microsoft mit meta-ai-fair benchmarking data-contamination multimodality fine-tuning ai-regulation ai-safety ai-weapons neural-networks model-architecture model-training model-performance robotics activation-functions long-context sam-altman jim-fan
Scale AI highlighted issues with data contamination in benchmarks like MMLU and GSM8K, proposing a new benchmark where Mistral overfits and Phi-3 performs well. Reka released the VibeEval benchmark for multimodal models addressing multiple choice benchmark limitations. Sam Altman of OpenAI discussed GPT-4 as "dumb" and hinted at GPT-5 with AI agents as a major breakthrough. Researchers jailbroke GPT-3.5 via fine-tuning. Global calls emerged to ban AI-powered weapons, with US officials urging human control over nuclear arms. Ukraine launched an AI consular avatar, while Moderna partnered with OpenAI for medical AI advancements. Sanctuary AI and Microsoft collaborate on AI for general-purpose robots. MIT introduced Kolmogorov-Arnold networks with improved neural network efficiency. Meta AI is training Llama 3 models with over 400 billion parameters, featuring multimodality and longer context.
Mergestral, Meta MTIAv2, Cohere Rerank 3, Google Infini-Attention
mistral-8x22b command-r-plus rerank-3 infini-attention llama-3 sd-1.5 cosxl meta-ai-fair mistral-ai cohere google stability-ai hugging-face ollama model-merging training-accelerators retrieval-augmented-generation linear-attention long-context foundation-models image-generation rag-pipelines model-benchmarking context-length model-performance aidan_gomez ylecun swyx
Meta announced their new MTIAv2 chips designed for training and inference acceleration with improved architecture and integration with PyTorch 2.0. Mistral released the 8x22B Mixtral model, which was merged back into a dense model to effectively create a 22B Mistral model. Cohere launched Rerank 3, a foundation model enhancing enterprise search and retrieval-augmented generation (RAG) systems supporting 100+ languages. Google published a paper on Infini-attention, an ultra-scalable linear attention mechanism demonstrated on 1B and 8B models with 1 million sequence length. Additionally, Meta's Llama 3 is expected to start rolling out soon. Other notable updates include Command R+, an open model surpassing GPT-4 in chatbot performance with 128k context length, and advancements in Stable Diffusion models and RAG pipelines.
Claude 3 is officially America's Next Top Model
claude-3-opus claude-3-sonnet claude-3-haiku gpt-4o-mini mistral-7b qwen-72b anthropic mistral-ai huggingface openrouter stable-diffusion automatic1111 comfyui fine-tuning model-merging alignment ai-ethics benchmarking model-performance long-context cost-efficiency model-evaluation mark_riedl ethanjperez stuhlmueller ylecun aravsrinivas
Claude 3 Opus outperforms GPT4T and Mistral Large in blind Elo rankings, with Claude 3 Haiku marking a new cost-performance frontier. Fine-tuning techniques like QLoRA on Mistral 7B and evolutionary model merging on HuggingFace models are highlighted. Public opinion shows strong opposition to ASI development. Research supervision opportunities in AI alignment are announced. The Stable Diffusion 3 (SD3) release raises workflow concerns for tools like ComfyUI and automatic1111. Opus shows a 5% performance dip on OpenRouter compared to the Anthropic API. A new benchmark stresses LLM recall at long contexts, with Mistral 7B struggling and Qwen 72b performing well.
Claude 3 just destroyed GPT 4 (see for yourself)
claude-3 claude-3-opus claude-3-sonnet claude-3-haiku gpt-4 anthropic amazon google claude-ai multimodality vision long-context model-alignment model-evaluation synthetic-data structured-output instruction-following model-speed cost-efficiency benchmarking safety mmitchell connor-leahy
Claude 3 from Anthropic launches in three sizes: Haiku (small, unreleased), Sonnet (medium, default on claude.ai, AWS, and GCP), and Opus (large, on Claude Pro). Opus outperforms GPT-4 on key benchmarks like GPQA, impressing benchmark authors. All models support multimodality with advanced vision capabilities, including converting a 2-hour video into a blog post. Claude 3 offers improved alignment, fewer refusals, and extended context length up to 1 million tokens with near-perfect recall. Haiku is noted for speed and cost-efficiency, processing dense research papers in under three seconds. The models excel at following complex instructions and producing structured outputs like JSON. Safety improvements reduce refusal rates, though some criticism remains from experts. Claude 3 is trained on synthetic data and shows strong domain-specific evaluation results in finance, medicine, and philosophy.
Ring Attention for >1M Context
gemini-pro gemma-7b gemma-2b deepseek-coder-6.7b-instruct llama-cpp google cuda-mode nvidia polymind deepseek ollama runpod lmstudio long-context ringattention pytorch cuda llm-guessing-game chatbots retrieval-augmented-generation vram-optimization fine-tuning dynamic-prompt-optimization ml-workflows gpu-scaling model-updates liu zaharia abbeel
Google Gemini Pro has sparked renewed interest in long context capabilities. The CUDA MODE Discord is actively working on implementing the RingAttention paper by Liu, Zaharia, and Abbeel, including extensions from the World Model RingAttention paper, with available PyTorch and CUDA implementations. TheBloke Discord discussed various topics including LLM guessing game evaluation, chatbot UX comparisons between Nvidia's Chat with RTX and Polymind, challenges in retrieval-augmented generation (RAG) integration, VRAM optimization, fine-tuning for character roleplay using Dynamic Prompt Optimization (DPO), and model choices like deepseek-coder-6.7B-instruct. There was also discussion on ML workflows on Mac Studio, with preferences for llama.cpp over ollama, and scaling inference cost-effectively using GPUs like the 4090 on Runpod. LM Studio users face manual update requirements for version 0.2.16, which includes support for Gemma models and bug fixes, especially for MacOS. The Gemma 7B model has had performance issues, while Gemma 2B received positive feedback.
Google AI: Win some (Gemma, 1.5 Pro), Lose some (Image gen)
gemma-2b gemma-7b gemma gemini-pro-1.5 llama-2 llama-3 mistral google hugging-face nvidia benchmarking license-policies image-generation video-understanding long-context dataset-editing model-integration gpu-hardware bug-fixes quantization
Google's Gemma open models (2-7B parameters) outperform Llama 2 and Mistral in benchmarks but face criticism for an unusual license and poor image generation quality, which Google partially acknowledges. The upcoming Gemini Pro 1.5 model features a 1 million token context window, excelling in video understanding and needle-in-haystack tasks. Discord communities like TheBloke and LM Studio discuss mixed reception of Gemma models, anticipation for Llama 3 release, challenges in dataset editing, and hardware considerations such as NVIDIA GeForce RTX 3090 and RTX 4090 GPUs. LM Studio users report issues with version 0.2.15 Beta and ongoing integration of Gemma models, with resources shared on Hugging Face.
Sora pushes SOTA
gemini-1.5 sora h20-gpt mistral-7b llama-13b mistralcasualml mixtral-instruct yi-models openai google-deepmind nvidia mistral-ai h2oai multimodality gpu-power-management long-context model-merging fine-tuning retrieval-augmented-generation role-play-model-optimization cross-language-integration training-loss synthetic-data-generation coding-support
Discord communities analyzed over 20 guilds, 312 channels, and 10550 messages reveal intense discussions on AI developments. Key highlights include the Dungeon Master AI assistant for Dungeons and Dragons using models like H20 GPT, GPU power supply debates involving 3090 and 3060 GPUs, and excitement around Google's Gemini 1.5 with its 1 million token context window and OpenAI's Sora model. Challenges with large world models (LWM) multimodality, GPT-assisted coding, and role-play model optimization with Yi models and Mixtral Instruct were discussed. Technical issues like model merging errors with MistralCasualML, fine-tuning scripts like AutoFineTune, and cross-language engineering via JSPyBridge were also prominent. NVIDIA's Chat with RTX feature leveraging retrieval-augmented generation (RAG) on 30+ series GPUs was compared to LMStudio's support for Mistral 7b and Llama 13b models. The community is cautiously optimistic about these frontier models' applications in media and coding.
1/8/2024: The Four Wars of the AI Stack
mixtral mistral nous-research openai mistral-ai hugging-face context-window distributed-models long-context hierarchical-embeddings agentic-rag fine-tuning synthetic-data oil-and-gas embedding-datasets mixture-of-experts model-comparison
The Nous Research AI Discord discussions highlighted several key topics including the use of DINO, CLIP, and CNNs in the Obsidian Project. A research paper on distributed models like DistAttention and DistKV-LLM was shared to address cloud-based LLM service challenges. Another paper titled 'Self-Extend LLM Context Window Without Tuning' argued that existing LLMs can handle long contexts inherently. The community also discussed AI models like Mixtral, favored for its 32k context window, and compared it with Mistral and Marcoroni. Other topics included hierarchical embeddings, agentic retrieval-augmented generation (RAG), synthetic data for fine-tuning, and the application of LLMs in the oil & gas industry. The launch of the AgentSearch-V1 dataset with one billion embedding vectors was also announced. The discussions covered mixture-of-experts (MoE) implementations and the performance of smaller models.
12/8/2023 - Mamba v Mistral v Hyena
mistral-8x7b-moe mamba-3b stripedhyena-7b claude-2.1 gemini gpt-4 dialogrpt-human-vs-machine cybertron-7b-v2-gguf falcon-180b mistral-ai togethercompute stanford anthropic google hugging-face mixture-of-experts attention-mechanisms prompt-engineering alignment image-training model-deployment gpu-requirements cpu-performance model-inference long-context model-evaluation open-source chatbots andrej-karpathy tri-dao maxwellandrews raddka
Three new AI models are highlighted: Mistral's 8x7B MoE model (Mixtral), Mamba models up to 3B by Together, and StripedHyena 7B, a competitive subquadratic attention model from Stanford's Hazy Research. Discussions on Anthropic's Claude 2.1 focus on its prompting technique and alignment challenges. The Gemini AI from Google is noted as potentially superior to GPT-4. The community also explores Dreambooth for image training and shares resources like the DialogRPT-human-vs-machine model on Hugging Face. Deployment challenges for large language models, including CPU performance and GPU requirements, are discussed with references to Falcon 180B and transformer batching techniques. User engagement includes meme sharing and humor.