All tags
Model: "qwen3-vl"
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.
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 doesn’t have to mean MoE routers + expert parallelism; static sparsity can be systems-friendly."
MCP -> Agentic AI Foundation, Mistral Devstral 2
devstral-2 devstral-small-2 sonnet-4.3 deepseek-v3.2 qwen3-vl openai anthropic block mistral-ai alibaba linux-foundation deepseek agentic-ai coding-models reinforcement-learning model-performance model-optimization open-weights cli-tools multi-file-code-automation data-decontamination moe reward-models rl-stability guillaumelample b_roziere qtnx_ charliermarsh omarsar0 eliebakouch justinwaugh cwolferesearch pan
OpenAI Engineering sees a significant collaborative milestone with the launch of the Agentic AI Foundation under the Linux Foundation, uniting projects from Anthropic, OpenAI, and Block. Mistral released Devstral 2, a coding model with 123B parameters and open weights, offering a cost-effective alternative to Sonnet 4.3 and competitive performance against DeepSeek v3.2. The new Mistral Vibe CLI supports agentic coding workflows with rapid ecosystem integration. Alibaba introduced Soft Adaptive Policy Optimization (SAPO) for reinforcement learning tuning, improving stability and performance in Qwen3-VL across multiple tasks. Research highlights include the importance of data decontamination in RL and ongoing discussions on MoE RL stability and reward hacking mitigation.
not much happened today
vllm-0.12.0 gemma3n qwen3-omni qwen3-vl gpt-5.1-codex-max gemini-3-pro runway-gen-4.5 kling-video-2.6 vllm nvidia huggingface langchain-ai together-ai meta-ai-fair sonarsource openrouter runway gemini arena gpu-programming quantization multimodality agent-platforms reinforcement-learning static-analysis reasoning inference-infrastructure model-optimization economics audio video-generation jeremyphoward mervenoyann sydneyrunkle swyx maximelabonne
vLLM 0.12.0 introduces DeepSeek support, GPU Model Runner V2, and quantization improvements with PyTorch 2.9.0 and CUDA 12.9. NVIDIA launches CUDA Tile IR and cuTile Python for advanced GPU tensor operations targeting Blackwell GPUs. Hugging Face releases Transformers v5 RC with an any-to-any multimodal pipeline supporting models like Gemma3n and Qwen3-Omni. Agent platforms see updates from LangChain with content moderation and cost tracking, Together AI and Meta AI collaborate on RL for long-horizon workflows, and SonarSource integrates static analysis into AI codegen. Economic insights from OpenRouter highlight coding as a key AI application, with reasoning models surpassing 50% usage and market bifurcation between premium and open models. Additionally, Kling Video 2.6 debuts native audio capabilities, and Runway Gen-4.5, Qwen3-TTS, and Gemini 3 Pro advance multimodality.
not much happened today
qwen3-max qwen3-vl qwen3-coder-plus gpt-5-codex code-world-model-32b claude-sonnet-4 claude-opus-4.1 alibaba openai meta-ai-fair huggingface anthropic microsoft github context-windows code-generation model-releases model-benchmarking api model-optimization multimodality software-engineering model-training huybery akhaliq lmarena_ai gdb ylecun pierceboggan julesagent
Alibaba unveiled the Qwen3 model family including Qwen3-Max and Qwen3-VL with a native 256K context window expandable to 1M, strong OCR in 32 languages, and rapid release velocity (~3.5 releases/month) backed by a $52B infrastructure roadmap. OpenAI launched GPT-5 Codex, an agent-optimized coding model with up to 400K context and adaptive reasoning priced at $1.25/$10 per million tokens, integrated into Cline and benchmarked in WebDev arenas. Meta AI FAIR released the open-weight Code World Model (CWM) 32B, a dense code generation model with strong benchmark scores (e.g., 65.8% SWE-bench Verified, 96.6% Math-500) and public safety reports. Ecosystem updates include GitHub Copilot's new embedding model for faster code search and Anthropic's Claude Sonnet 4 and Opus 4.1 integration into Microsoft 365 Copilot. The vLLM 0.10.2 update introduces Decode Context Parallel (DCP) for improved system performance.
Alibaba Yunqi: 7 models released in 4 days (Qwen3-Max, Qwen3-Omni, Qwen3-VL) and $52B roadmap
qwen3-max qwen3-omni qwen3-vl qwen3guard qwen3-livetranslate qwen3-tts-flash qwen-image-edit qwen3coder qwen alibaba alicloud tool-use large-model-coding reasoning multimodality model-release model-updates industry-application scaling fine-tuning reinforcement-learning junyang_lin eddie_wu alibaba_wan
Alibaba's Tongyi Qianwen (Qwen) team launched major updates including the 1T parameter Qwen3-Max, Qwen3-Omni, and Qwen3-VL models, alongside specialized versions like Qwen3Guard, Qwen3-LiveTranslate, Qwen3-TTS-Flash, Qwen-Image-Edit, and Qwen3Coder. At the AliCloud Yunqi (Apsara) conference, CEO Eddie Wu outlined a $52B roadmap emphasizing two AI development stages: "intelligence emergence" focusing on learning from humans and reasoning, and "autonomous action" highlighting AI's tool use and real-world task execution. The updates showcase advances in tool use, large-model coding capabilities, and AI's expanding role across industries such as logistics, manufacturing, biomedicine, and finance. Junyang Lin and Alibaba Wan are key spokespersons for these developments. The Qwen project is now seen as a "frontier lab" for AI innovation.