←Search results

Search “Reasoning”

By relevance · Total 3279 results Clear search

Sunday

Saturday

Friday

  1. Tongyi Qianwen / QwenSelected

    Alibaba Qwen Releases Omni-Modal Model Qwen3.8-Omni-Flash

    Alibaba Qwen has released Qwen3.8-Omni-Flash, its first omni-modal model built around agent capabilities, with native support for audio-visual understanding, reasoning, and tool calling. The company says its audio-visual capabilities are close to Gemini 3.8 Flash, with agent performance improving by an average of 19.5 points on WildClawBench-MM and UniClawBench; it supports a 1M token context and uses 51.8% fewer tokens than static understanding on OmniVideoBench. Video input costs are about 89% lower than Qwen3.5-Omni-Plus, while Qwen-MM-Plugins is being open-sourced and Qwen-Live Harness is coming soon.

    Alibaba Qwen has released Qwen3.8-Omni-Flash, its first omni-modal model built around agent capabilities, with native support for audio-visual understanding, reasoning, and tool calling. The company says its audio-visual capabilities are close to Gemini 3.8 Flash, with agent performance improving by an average of 19.5 points on WildClawBench-MM and UniClawBench; it supports a 1M token context and uses 51.8% fewer tokens than static understanding on OmniVideoBench. Video input costs are about 89%

    Matched fieldsSummary

    #company#Model releases#Multimodal#Agents

Thursday

Wednesday

  3. Baseten Base Labs: Model Research

    Baseten Partners with Goodfire and Hugging Face to Build Open-Source Model Safety Evaluation and Monitoring Infrastructure

    Baseten and Base Labs announced a joint effort with Goodfire and Hugging Face to build safety evaluation and monitoring infrastructure for the open-source models they serve, integrating safety research directly into training and inference systems. The work covers three stages: training (getting models to follow explicit policies, recognize capability boundaries, and proactively ask for help), monitoring and intervention (studying inputs, outputs, tool calls, and model internal states, and connecting signals to permissions, isolation, and pre-execution checks), and learning from failures. Base Labs is responsible for developing and publishing the methods, while Baseten integrates them into its deployment infrastructure, offering them as a hosted service and publishing the methods and evidence so customers and external evaluators can verify them independently.

    Charles

    Baseten and Base Labs announced a joint effort with Goodfire and Hugging Face to build safety evaluation and monitoring infrastructure for the open-source models they serve, integrating safety research directly into training and inference systems. The work covers three stages: training (getting models to follow explicit policies, recognize capability boundaries, and proactively ask for help), monitoring and intervention (studying inputs, outputs, tool calls, and model internal states, and connec

    Matched fieldsSummary

    #lab#Safety / alignment#Open-source ecosystem#English text is being prepared.