All tags
Person: "_akhaliq"
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 <10× 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.
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.
Grok 4 Fast: Xai's distilled, 40% more token efficient, 2m context, 344 tok/s frontier model
grok-4-fast magistral-1.2 moondream-3 granite-docling-258m sail-vl2 xai meta-ai-fair mistral-ai ibm bytedance efficiency reasoning vision multimodality model-optimization model-deployment vision-encoders model-architecture model-training nearcyan aidangomez _akhaliq vikhyatk rohanpaul_ai
xAI announced Grok 4 Fast, a highly efficient model running at 344 tokens/second, offering reasoning and nonreasoning modes and free trials on major platforms. Meta showcased its neural band and Ray-Ban Display with a live demo that experienced hiccups but sparked discussion on live hardware demos and integration challenges. Meta is also developing a first-party "Horizon Engine" for AI rendering and released Quest-native Gaussian Splatting capture tech. New model releases include Mistral's Magistral 1.2, a compact multimodal vision-language model with improved benchmarks and local deployment; Moondream 3, a 9B-parameter MoE VLM focused on efficient visual reasoning; IBM's Granite-Docling-258M, a document VLM for layout-faithful PDF to HTML/Markdown conversion; and ByteDance's SAIL-VL2, a vision-language foundation model excelling at multimodal understanding and reasoning at 2B and 8B parameter scales.
Softbank, NVIDIA and US Govt take 2%, 5% and 10% of Intel, will develop Intel x86 RTX SOCs for consumer & datacenters
magistral-1.2 moondream-3 nvidia intel meta-ai-fair mistral-ai multimodality vision model-optimization model-efficiency model-architecture reinforcement-learning fine-tuning ai-hardware gaussian-splatting live-demo visual-reasoning nearcyan _akhaliq vikhyatk
Nvidia and Intel announced a joint development partnership for multiple new generations of x86 products, marking a significant shift in the tech industry. This collaboration has been in the works for a year and impacts both consumer and data center markets, boosting hopes for Intel's Foundry business. On the AI hardware front, Meta showcased its neural band and Ray-Ban Display with a live demo that experienced hiccups but sparked discussion on live tech demos. Meta is also moving from Unity to its own Horizon Engine for AI rendering, including Gaussian splatting capture technology. In AI models, Mistral released Magistral 1.2, a compact multimodal vision-language model with improved benchmarks and local deployment capabilities, while Moondream 3 previewed a 9B-parameter, 2B-active MoE VLM focused on efficient visual reasoning.
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.
not much happened today
embeddinggemma qwen-2.5-coder minicpm-v-4.5 gpt-4o gemini-2.0-pro google-deepmind hugging-face jina-ai lighton microsoft stanford openai ollama weaviate langchain llamaindex embeddings retrieval-augmented-generation quantization multilingual-models on-device-ai semantic-search contrastive-learning dataset-release vision multimodality video-generation text-to-speech optimizer-benchmarking training-recipes model-compression video-token-compression fine-tuning osanseviero _philschmid tomaarsen ollama weaviate_io lusxvr andimarafioti thibaudfrere _akhaliq clementdelangue gordonwetzstein konstmish wen_kaiyue percyliang
Google DeepMind released EmbeddingGemma (308M), a small multilingual embedding model optimized for on-device retrieval-augmented generation and semantic search, supporting over 100 languages and running efficiently with quantization and EdgeTPU latency under 15ms. Jina AI introduced new code-focused embedding models (0.5B/1.5B) with GGUF quantization, achieving state-of-the-art retrieval across multiple languages and tasks. LightOn demonstrated large-scale retrieval training without distillation using contrastive training on billions of passages. Hugging Face released the FineVision dataset with 17.3M images and 9.5B answer tokens for vision-language model training, showing significant benchmark improvements. The MiniCPM-V 4.5 (8B) multimodal model reported surpassing GPT-4o and Gemini-2.0 Pro on OpenCompass benchmarks with innovative video token compression. Microsoft’s VibeVoice TTS and Stanford’s Mixture-of-Contexts video generation also featured. Additionally, a Stanford study benchmarked optimizers like Muon, Soap, Mars, and Sophia, finding diminishing speedups over AdamW at larger scales but advantages at smaller scales. The new ChatGPT branching feature was noted for its simplicity and popularity. "Everyone's a decacorn now."
not much happened today
claude-code gemini qwen3-coder gemini-2.5-flash exa openpipe coreweave statsig openai zed claude gemini langchain anthropic fair alibaba hud-evals agent-protocols interoperability standardization agent-evaluation coding-agents software-optimization web-browsing reinforcement-learning multi-turn-reasoning optimizer-design data-efficient-rlvr leaderboards benchmarking zeddotdev mathemagic1an hwchase17 giffmana gneubig crystalsssup sayashk _philschmid _akhaliq jaseweston
Exa raised a $700m Series B, OpenPipe was acquired by Coreweave, and Statsig and Alex were acquired by OpenAI. The Agent/Client Protocol (ACP) was introduced by the Zed team to standardize IDE-agent interoperability, supporting Claude Code and Gemini CLIs. LangChain 1.0 alpha unifies content blocks for reasoning and multimodal data. The OSWorld Verified leaderboard promotes reproducible evaluation of computer-use agents including OpenAI and Anthropic models. FAIR revealed coding agent cheating on SWE-Bench Verified. PR Arena hosts live coding agent competitions. Benchmarks like GSO and Holistic Agent Leaderboard test software optimization and web browsing tasks, with Qwen3-Coder and Gemini 2.5 Flash showing strong performance. Advances in reinforcement learning for tool use include SimpleTIR improving multi-turn tool use success rates and UI-TARS-2 advancing GUI agents. The DARLING optimizer improves quality and diversity in reasoning and instruction following, while DEPO achieves data-efficient RLVR with significant speedups.
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.
not much happened today
gemma-3-270m canary-1b parakeet-tdt-0.6b nemotron-nano-v2 qwen-image-edit dino-v3 nvidia alibaba tencent meta-ai-fair ibm datology synthetic-data multilingual-asr self-supervised-learning vision model-efficiency training-data data-augmentation model-speedup domain-transfer demishassabis adrgrondin rasbt reach_vb ctnzr clementdelangue natolambert _akhaliq itspaulai mervenoyann xenovacom tomaarsen pratyushmaini code_star leavittron k_schuerholt giffmana
Gemma 3 270M, an ultra-small model optimized for edge and mobile use, was released and is gaining adoption. NVIDIA launched two open multilingual ASR models, Canary 1B and Parakeet-TDT 0.6B, trained on 1 million hours of data with CC-BY licensing, plus the efficient Nemotron-Nano v2 9B model with significant speedups. Alibaba's Qwen-Image-Edit offers bilingual text editing and semantic image transformations. Tencent Hunyuan introduced a controllable game-world video generator trained on over 1 million gameplay recordings. Meta's DINOv3 presents a scalable self-supervised vision backbone with strong domain transfer capabilities. IBM quietly released efficient English embedding models under a commercial-friendly license. The BeyondWeb synthetic data paper shows significant training speed and performance gains over prior datasets. Analysis of HRM architecture suggests performance improvements largely stem from data augmentation and scaffolding rather than novel architecture. "Models and datasets are openly licensed and available on Hugging Face."
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.
not much happened today
grok-4 jamba ernie-4.5 claude-4-sonnet claude-4 kontext-dev ai21-labs hugging-face baidu perplexity-ai deepmind anthropic reinforcement-learning fine-tuning energy-based-transformers ssm-transformer context-windows length-generalization recurrent-neural-networks attention-mechanisms 2-simplicial-attention biomedical-ai instruction-following open-weight-models python-package-management _philschmid corbtt jxmnop sedielem _akhaliq slashml alexiglad clementdelangue _albertgu tri_dao theaitimeline deep-learning-ai
Over the holiday weekend, key AI developments include the upcoming release of Grok 4, Perplexity teasing new projects, and community reactions to Cursor and Dia. Research highlights feature a paper on Reinforcement Learning (RL) improving generalization and reasoning across domains, contrasting with Supervised Fine-Tuning's forgetting issues. Energy-Based Transformers (EBTs) are proposed as a promising alternative to traditional transformers. AI21 Labs updated its Jamba model family with enhanced grounding and instruction following, maintaining a 256K context window. Baidu open-sourced its massive 424 billion parameter Ernie 4.5 model, while Kontext-dev became the top trending model on Hugging Face. Advances in length generalization for recurrent models and the introduction of 2-simplicial attention were noted. In biomedical AI, Biomni, powered by Claude 4 Sonnet, demonstrated superior accuracy and rare disease diagnosis capabilities. Additionally, the Python package manager
uv received praise for improving Python installation workflows. 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.
not much happened today
codex claude-4-opus claude-4-sonnet gemini-2.5-pro gemini-2.5 qwen-2.5-vl qwen-3 playdiffusion openai anthropic google perplexity-ai bing playai suno hugging-face langchain-ai qwen mlx assemblyai llamacloud fine-tuning model-benchmarking text-to-video agentic-ai retrieval-augmented-generation open-source-models speech-editing audio-processing text-to-speech ultra-low-latency multimodality public-notebooks sama gdb kevinweil lmarena_ai epochairesearch reach_vb wightmanr deeplearningai mervenoyann awnihannun jordirib1 aravsrinivas omarsar0 lioronai jerryjliu0 nerdai tonywu_71 _akhaliq clementdelangue _mfelfel
OpenAI rolled out Codex to ChatGPT Plus users with internet access and fine-grained controls, improving memory features for free users. Anthropic's Claude 4 Opus and Sonnet models lead coding benchmarks, while Google's Gemini 2.5 Pro and Flash models gain recognition with new audio capabilities. Qwen 2.5-VL and Qwen 3 quantizations are noted for versatility and support. Bing Video Creator launched globally enabling text-to-video generation, and Perplexity Labs sees increased demand for travel search. New agentic AI tools and RAG innovations include LlamaCloud and FedRAG. Open-source releases include Holo-1 for web navigation and PlayAI's PlayDiffusion for speech editing. Audio and multimodal advances feature Suno's music editing upgrades, Google's native TTS in 24+ languages, and Universal Streaming's ultra-low latency speech-to-text. Google NotebookLM now supports public notebooks. "Codex's internet access brings tradeoffs, with explicit warnings about risk" and "Gemini 2.5 Pro is cited as a daily driver by users".
Mary Meeker is so back: BOND Capital AI Trends report
qwen-3-8b anthropic hugging-face deepseek attention-mechanisms inference arithmetic-intensity transformers model-optimization interpretability model-quantization training tri_dao fleetwood___ teortaxestex awnihannun lateinteraction neelnanda5 eliebakouch _akhaliq
Mary Meeker returns with a comprehensive 340-slide report on the state of AI, highlighting accelerating tech cycles, compute growth, and comparisons of ChatGPT to early Google and other iconic tech products. The report also covers enterprise traction and valuation of major AI companies. On Twitter, @tri_dao discusses an "ideal" inference architecture featuring attention variants like GTA, GLA, and DeepSeek MLA with high arithmetic intensity (~256), improving efficiency and model quality. Other highlights include the release of 4-bit DWQ of DSR1 Qwen3 8B on Hugging Face, AnthropicAI's open-source interpretability tools for LLMs, and discussions on transformer training and abstractions by various researchers.
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.
Granola launches team notes, while Notion launches meeting transcription
gpt-4.1 gpt-4o-mini gpt-4.1-mini claude-opus claude-sonnet claude-o3 qwen3 seed1.5-vl llama-4 am-thinking-v1 openai anthropic alibaba meta-ai-fair huggingface granola coding instruction-following benchmarking model-releases reasoning image-generation collaborative-software model-performance kevinweil scaling01 steph_palazzolo andersonbcdefg reach_vb yuchenj_uw qtnx_ _akhaliq risingsayak
GPT-4.1 is now available in ChatGPT for Plus, Pro, and Team users, focusing on coding and instruction following, with GPT 4.1 mini replacing GPT 4o mini. Anthropic is releasing new Claude models including Claude Opus and Claude Sonnet, though some criticism about hallucinations in Claude O3 was noted. Alibaba shared the Qwen3 Technical Report with strong benchmark results from Seed1.5-VL. Meta FAIR announced new models and datasets but faced criticism on Llama 4. AM-Thinking-v1 launched on Hugging Face as a 32B scale reasoning model. Granola raised $43M in Series B and launched Granola 2.0 with a Notion-like UI. The AI ecosystem shows rapid iteration and cloning of ideas, emphasizing execution and distribution.
not much happened today
hunyuan-turbos qwen3-235b-a22b o3 gpt-4.1-nano grok-3 gemini-2.5-pro seed1.5-vl kling-2.0 tencent openai bytedance meta-ai-fair nvidia deepseek benchmarking model-performance moe reasoning vision video-understanding vision-language multimodality model-evaluation model-optimization lmarena_ai artificialanlys gdb _jasonwei iScienceLuvr _akhaliq _philschmid teortaxesTex mervenoyann reach_vb
Tencent's Hunyuan-Turbos has risen to #8 on the LMArena leaderboard, showing strong performance across major categories and significant improvement since February. The Qwen3 model family, especially the Qwen3 235B-A22B (Reasoning) model, is noted for its intelligence and efficient parameter usage. OpenAI introduced HealthBench, a new health evaluation benchmark developed with input from over 250 physicians, where models like o3, GPT-4.1 nano, and Grok 3 showed strong results. ByteDance released Seed1.5-VL, a vision-language model with a 532M-parameter vision encoder and a 20B active parameter MoE LLM, achieving state-of-the-art results on 38 public benchmarks. In vision-language, Kling 2.0 leads image-to-video generation, and Gemini 2.5 Pro excels in video understanding with advanced multimodal capabilities. Meta's Vision-Language-Action framework and updates on VLMs for 2025 were also highlighted.
Prime Intellect's INTELLECT-2 and PRIME-RL advance distributed reinforcement learning
intellect-2 dreamo qwen gemini-2.5-pro dynamic-byte-latent-transformer gen-4-references mistral-medium-3 le-chat-enterprise primeintellect bytedance qwen gemma meta-ai-fair runwayml mistral-ai google distributed-training reinforcement-learning gpu-clusters model-optimization quantization multimodality agentic-ai video-understanding fine-tuning _akhaliq reach_vb osanseviero aiatmeta c_valenzuelab lmarena_ai adcock_brett
Prime Intellect released INTELLECT-2, a decentralized GPU training and RL framework with a vision for distributed AI training overcoming colocation limits. ByteDance launched DreamO, a unified image customization model on Hugging Face. Qwen released models optimized for GPTQ, GGUF, and AWQ quantization. Gemma surpassed 150 million downloads on Hugging Face. Meta released weights for the Dynamic Byte Latent Transformer and the Collaborative Reasoner framework to improve language model efficiency and reasoning. RunwayML introduced Gen-4 References, a near-realtime model requiring no fine-tuning. Mistral AI released Mistral Medium 3, a strong multimodal model, and Le Chat Enterprise, an agentic AI assistant for business. Google updated Gemini 2.5 Pro Preview with video understanding and UI improvements. "Airbnb for spare GPUs from all over the world" highlights the ongoing challenges and potential of distributed GPU training.
Cursor @ $9b, OpenAI Buys Windsurf @ $3b
llama-nemotron-ultra llama-nemotron-super llama-nemotron-nano qwen3-235b-a22b prover-v2 phi-4-reasoning ernie-4.5-turbo ernie-x1-turbo suno-v4.5 gen-4-references o1-mini openai cursor nvidia alibaba deepseek microsoft baidu suno runway keras reasoning inference-efficiency open-license moe-models math-reasoning theorem-proving model-performance music-generation image-generation recommender-systems tpu-optimization _akhaliq adcock_brett lmarena_ai fchollet
OpenAI is reportedly close to closing a deal with Windsurf, coinciding with Cursor's $900M funding round at a $9B valuation. Nvidia launched the Llama-Nemotron series featuring models from 8B to 253B parameters, praised for reasoning and inference efficiency. Alibaba released the Qwen3 family with MoE and dense models up to 235B parameters, ranking highly in coding and math benchmarks. DeepSeek introduced Prover-V2, an open-source AI for math reasoning with an 88.9% pass rate on MiniF2F-test. Microsoft released reasoning-focused Phi-4 models, outperforming OpenAI's o1-mini. Baidu debuted turbo versions of ERNIE 4.5 and X1 for faster, cheaper inference. Suno v4.5 added advanced AI music generation features, while Runway Gen-4 References enable placing characters into scenes with high consistency. KerasRS, a new recommender system library optimized for TPUs, was released by Fran ois Chollet.
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.
Google's Agent2Agent Protocol (A2A)
kimi-vl-a3b gpt-4o llama-4-scout llama-4-maverick llama-4-behemoth deepcoder-14b o3-mini o1 llama-3.1-nemotron-ultra-253b deepseek-r1 google google-deepmind moonshot-ai meta-ai-fair uc-berkeley openai nvidia hugging-face togethercompute deepseek agent-interoperability multimodality vision math reinforcement-learning coding model-training open-source model-benchmarking context-windows streaming push-notifications enterprise-authentication model-release reach_vb _akhaliq epochairesearch artificialanlys winglian danielhanchen yuchenj_uw jeremyphoward
Google Cloud Next announcements featured the launch of Google and DeepMind's full MCP support and a new Agent to Agent protocol designed for agent interoperability with multiple partners. The protocol includes components like the Agent Card, Task communication channels, Enterprise Auth and Observability, and Streaming and Push Notification support. On the model front, Moonshot AI released Kimi-VL-A3B, a multimodal model with 128K context and strong vision and math benchmark performance, outperforming gpt-4o. Meta AI introduced smaller versions of llama-4 family models: llama-4-scout and llama-4-maverick, with a larger Behemoth model still in training. DeepCoder 14B from UC Berkeley is an open-source coding model rivaling openai's o3-mini and o1 models, trained with reinforcement learning on 24K coding problems. Nvidia released llama-3.1-nemotron-ultra-253b on Hugging Face, noted for beating llama-4-behemoth and maverick and competing with deepseek-r1.
DeepSeek's Open Source Stack
qwen-qwq-32b start character-3 gemini gemini-2.0 mercury-coder gpt-4.5 jamba-mini-1.6 gemini-2.0-flash gpt-4o-mini mistral-small-3 mistral-ocr deepseek pyspur hugging-face togethercompute hedra-labs google-deepmind deeplearningai openai ai21-labs mistral-ai fine-tuning benchmarking multimodality code-generation diffusion-models model-performance model-optimization ocr embedding-models context-windows runtime-limits _akhaliq lmarena_ai reach_vb danielhanchen _philschmid aidan_mclau vikhyatk jerryjliu0
DeepSeek's Open Source Week was summarized by PySpur, highlighting multiple interesting releases. The Qwen QwQ-32B model was fine-tuned into START, excelling in PhD-level science QA and math benchmarks. Character-3, an omnimodal AI video generation model by Hedra Labs and Together AI, enables realistic animated content creation. Google DeepMind introduced the Gemini embedding model with an 8k context window, ranking #1 on MMTEB, alongside the Gemini 2.0 Code Executor supporting Python libraries and auto-fix features. Inception Labs' Mercury Coder is a diffusion-based code generation model offering faster token processing. OpenAI released GPT-4.5, their largest model yet but with less reasoning ability than some competitors. AI21 Labs launched Jamba Mini 1.6, noted for superior output speed compared to Gemini 2.0 Flash, GPT-4o mini, and Mistral Small 3. A new dataset of 1.9M scanned pages was released for OCR benchmarking, with Mistral OCR showing competitive but not top-tier document parsing performance compared to LLM/LVM-powered methods. "Cracked engineers are all you need."
not much happened today
claude-3.7-sonnet claude-3.7 deepseek-r1 o3-mini deepseek-v3 gemini-2.0-pro gpt-4o qwen2.5-coder-32b-instruct anthropic perplexity-ai amazon google-cloud deepseek_ai coding reasoning model-benchmarking agentic-workflows context-window model-performance open-source moe model-training communication-libraries fp8 nvlink rdma cli-tools skirano omarsar0 reach_vb artificialanlys terryyuezhuo _akhaliq _philschmid catherineols goodside danielhanchen
Claude 3.7 Sonnet demonstrates exceptional coding and reasoning capabilities, outperforming models like DeepSeek R1, O3-mini, and GPT-4o on benchmarks such as SciCode and LiveCodeBench. It is available on platforms including Perplexity Pro, Anthropic, Amazon Bedrock, and Google Cloud, with pricing at $3/$15 per million tokens. Key features include a 64k token thinking mode, 200k context window, and the CLI-based coding assistant Claude Code. Meanwhile, DeepSeek released DeepEP, an open-source communication library optimized for MoE model training and inference with support for NVLink, RDMA, and FP8. These updates highlight advancements in coding AI and efficient model training infrastructure.
The Ultra-Scale Playbook: Training LLMs on GPU Clusters
deepseek-native-sparse-attention r1-1776 paligemma-2-mix muse baichuan-m1-14b stripedhyena-2 huggingface deepseek perplexity-ai google-deepmind microsoft baichuan stripedhyena gpu-training scaling multimodality vision model-training foundation-models medical-llm genome-modeling robotic-manipulation interactive-content eliebakouch nouamanetazi lvwerra thom-wolf proftomyeh alex-wang aravsrinivas _akhaliq _philschmid mervenoyann reach_vb arankomatsuzaki maximelabonne
Huggingface released "The Ultra-Scale Playbook: Training LLMs on GPU Clusters," an interactive blogpost based on 4000 scaling experiments on up to 512 GPUs, providing detailed insights into modern GPU training strategies. DeepSeek introduced the Native Sparse Attention (NSA) model, gaining significant community attention, while Perplexity AI launched R1-1776, an uncensored and unbiased version of DeepSeek's R1 model. Google DeepMind unveiled PaliGemma 2 Mix, a multi-task vision-language model available in 3B, 10B, and 28B sizes. Microsoft introduced Muse, a generative AI model trained on the game Bleeding Edge, and presented Magma, a foundation model for multimodal AI agents excelling in UI navigation and robotic manipulation. Baichuan-M1-14B was announced as a state-of-the-art medical LLM trained on 20T tokens, and a fully open-source 40B genome modeling model using StripedHyena 2 architecture was also released. "Making your own gaming experience is coming sooner than you'd think," noted in relation to Muse.
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.
not much happened today
chatgpt-4o deepseek-r1 o3 o3-mini gemini-2-flash qwen-2.5 qwen-0.5b hugging-face openai perplexity-ai deepseek-ai gemini qwen metr_evals reasoning benchmarking model-performance prompt-engineering model-optimization model-deployment small-language-models mobile-ai ai-agents speed-optimization _akhaliq aravsrinivas lmarena_ai omarsar0 risingsayak
Smolagents library by Huggingface continues trending. ChatGPT-4o latest version
chatgpt-40-latest-20250129 released. DeepSeek R1 671B sets speed record at 198 t/s, fastest reasoning model, recommended with specific prompt settings. Perplexity Deep Research outperforms models like Gemini Thinking, o3-mini, and DeepSeek-R1 on Humanity's Last Exam benchmark with 21.1% score and 93.9% accuracy on SimpleQA. ChatGPT-4o ranks #1 on Arena leaderboard in multiple categories except math. OpenAI's o3 model powers Deep Research tool for ChatGPT Pro users. Gemini 2 Flash and Qwen 2.5 models support LLMGrading verifier. Qwen 2.5 models added to PocketPal app. MLX shows small LLMs like Qwen 0.5B generate tokens at high speed on M4 Max and iPhone 16 Pro. Gemini Flash 2.0 leads new AI agent leaderboard. DeepSeek R1 is most liked on Hugging Face with over 10 million downloads. not much happened today
claude-3.5-sonnet opencoder anthropic microsoft sambanova openai langchain llamaindex multi-agent-systems natural-language-interfaces batch-processing harmful-content-detection secret-management retrieval-augmented-generation error-analysis memory-management web-scraping autonomous-agents sophiamyang tom_doerr omarsar0 _akhaliq andrewyng giffmana
This week in AI news, Anthropic launched Claude Sonnet 3.5, enabling desktop app control via natural language. Microsoft introduced Magentic-One, a multi-agent system built on the AutoGen framework. OpenCoder was unveiled as an AI-powered code cookbook for large language models. SambaNova is sponsoring a hackathon with prizes up to $5000 for building real-time AI agents. Sophiamyang announced new Batch and Moderation APIs with 50% lower cost and multi-dimensional harmful text detection. Open-source tools like Infisical for secret management, CrewAI for autonomous agent orchestration, and Crawlee for web scraping were released. Research highlights include SCIPE for error analysis in LLM chains, Context Refinement Agent for improved retrieval-augmented generation, and MemGPT for managing LLM memory. The week also saw a legal win for OpenAI in the RawStory copyright case, affirming that facts used in LLM training are not copyrightable.
ChatGPT Advanced Voice Mode
o1-preview qwen-2.5 llama-3 claude-3.5 openai anthropic scale-ai togethercompute kyutai-labs voice-synthesis planning multilingual-datasets retrieval-augmented-generation open-source speech-assistants enterprise-ai price-cuts benchmarking model-performance sam-altman omarsar0 bindureddy rohanpaul_ai _philschmid alexandr_wang svpino ylecun _akhaliq
OpenAI rolled out ChatGPT Advanced Voice Mode with 5 new voices and improved accent and language support, available widely in the US. Ahead of rumored updates for Llama 3 and Claude 3.5, Gemini Pro saw a significant price cut aligning with the new intelligence frontier pricing. OpenAI's o1-preview model showed promising planning task performance with 52.8% accuracy on Randomized Mystery Blocksworld. Anthropic is rumored to release a new model, generating community excitement. Qwen 2.5 was released with models up to 32B parameters and support for 128K tokens, matching GPT-4 0613 benchmarks. Research highlights include PlanBench evaluation of o1-preview, OpenAI's release of a multilingual MMMLU dataset covering 14 languages, and RAGLAB framework standardizing Retrieval-Augmented Generation research. New AI tools include PDF2Audio for converting PDFs to audio, an open-source AI starter kit for local model deployment, and Moshi, a speech-based AI assistant from Kyutai. Industry updates feature Scale AI nearing $1B ARR with 4x YoY growth and Together Compute's enterprise platform offering faster inference and cost reductions. Insights from Sam Altman's blog post were also shared.
nothing much happened today
o1 chatgpt-4o llama-3-1-405b openai lmsys scale-ai cognition langchain qdrant rohanpaul_ai reinforcement-learning model-merging embedding-models toxicity-detection image-editing dependency-management automated-code-review visual-search benchmarking denny_zhou svpino alexandr_wang cwolferesearch rohanpaul_ai _akhaliq kylebrussell
OpenAI's o1 model faces skepticism about open-source replication due to its extreme restrictions and unique training advances like RL on CoT. ChatGPT-4o shows significant performance improvements across benchmarks. Llama-3.1-405b fp8 and bf16 versions perform similarly with cost benefits for fp8. A new open-source benchmark "Humanity's Last Exam" offers $500K in prizes to challenge LLMs. Model merging benefits from neural network sparsity and linear mode connectivity. Embedding-based toxic prompt detection achieves high accuracy with low compute. InstantDrag enables fast, optimization-free drag-based image editing. LangChain v0.3 releases with improved dependency management. Automated code review tool CodeRabbit adapts to team coding styles. Visual search advances integrate multimodal data for better product search. Experts predict AI will be default software by 2030.
Hybrid SSM/Transformers > Pure SSMs/Pure Transformers
mamba-2-hybrid gpt-4 qwen-72b table-llava-7b nvidia lamini-ai sakana-ai luma-labs mixture-of-experts benchmarking fine-tuning multimodality text-to-video model-performance memory-optimization preference-optimization video-understanding multimodal-tables bryan-catanzaro bindureddy ylecun ctnzr corbtt realsharonzhou andrew-n-carr karpathy _akhaliq omarsar0
NVIDIA's Bryan Catanzaro highlights a new paper on Mamba models, showing that mixing Mamba and Transformer blocks outperforms either alone, with optimal attention below 20%. Mixture-of-Agents (MoA) architecture improves LLM generation quality, scoring 65.1% on AlpacaEval 2.0 versus GPT-4 Omni's 57.5%. The LiveBench AI benchmark evaluates reasoning, coding, writing, and data analysis. A hybrid Mamba-2-Hybrid model with 7% attention surpasses a Transformer on MMLU accuracy, jumping from 50% to 53.6%. GPT-4 performs better at temperature=1. Qwen 72B leads open-source models on LiveBench AI. LaminiAI Memory Tuning achieves 95% accuracy on a SQL agent task, improving over instruction fine-tuning. Sakana AI Lab uses evolutionary strategies for preference optimization. Luma Labs Dream Machine demonstrates advanced text-to-video generation. The MMWorld benchmark evaluates multimodal video understanding, and Table-LLaVa 7B competes with GPT-4V on multimodal table tasks.
Mamba-2: State Space Duality
mamba-2 mamba transformer++ llama-3-70b gpt-3 hugging-face state-space-models perplexity training-efficiency data-pruning benchmarking multimodality video-analysis _albertgu tri_dao arankomatsuzaki _akhaliq clementdelangue karpathy
Mamba-2, a new state space model (SSM), outperforms previous models like Mamba and Transformer++ in perplexity and wall-clock time, featuring 8x larger states and 50% faster training. It introduces the concept of state space duality (SSD) connecting SSMs and linear attention. The FineWeb-Edu dataset, a high-quality subset of the 15 trillion token FineWeb dataset, filtered using llama-3-70b for educational quality, enables better and faster LLM learning, potentially reducing tokens needed to surpass GPT-3 performance. Additionally, perplexity-based data pruning using a 125M parameter model improves downstream performance and reduces pretraining steps by up to 1.45x. The Video-MME benchmark evaluates multi-modal LLMs on video analysis across multiple visual domains and video lengths.
The Era of 1-bit LLMs
bitnet-b1.58 hugging-face quantization model-optimization energy-efficiency fine-tuning robotics multimodality ai-security ethics humor swyx levelsio gdb npew _akhaliq osanseviero mmitchell_ai deliprao nearcyan clementdelangue
The Era of 1-bit LLMs research, including the BitNet b1.58 model, introduces a ternary parameter approach that matches full-precision Transformer LLMs in performance while drastically reducing energy costs by 38x. This innovation promises new scaling laws and hardware designs optimized for 1-bit LLMs. Discussions on AI Twitter highlight advances in AGI societal impact, robotics with multimodal models, fine-tuning techniques like ResLoRA, and AI security efforts at Hugging Face. Ethical considerations in generative AI and humor within the AI community are also prominent topics.