←Search results

Search “Reasoning”

By relevance · Total 3280 results Clear search

Tuesday

  7. UCSD Hao AI LabSelected

    UCSD Hao AI Lab open-sources Minecraft agent Jev, prompt evolution without training

    UCSD Hao AI Lab has open-sourced the Minecraft agent Jev, publishing how it works after 1,795 matches. It does not train a model; instead, a SOTA LLM agent evolves prompts from the bot's own logs under human supervision; in Block UHC it won 70% of rounds against human players. Each decision takes 24 ms on NVIDIA Blackwell, which the team says is 2.5 times faster than a mainstream inference engine, and it has open-sourced the inference engine DJev-serve, completing each decision in a single CUDA graph on GB200 NVL72 and B200.

    UCSD Hao AI Lab has open-sourced the Minecraft agent Jev, publishing how it works after 1,795 matches. It does not train a model; instead, a SOTA LLM agent evolves prompts from the bot's own logs under human supervision; in Block UHC it won 70% of rounds against human players. Each decision takes 24 ms on NVIDIA Blackwell, which the team says is 2.5 times faster than a mainstream inference engine, and it has open-sourced the inference engine DJev-serve, completing each decision in a single CUDA

    Matched fieldsSummary

    #lab#English text is being prepared.#Agents#English text is being prepared.
  13. Clément Delangue

    Reflection launches Beam, an open-source agentic model with 501B total parameters and 23B active

    Reflection has released Beam, an efficient open-source agentic model with 501B total parameters and 23B active parameters, emphasizing frontier reasoning efficiency and advancing the Western open-source frontier on coding and agentic tasks, trained end-to-end from scratch. Full weights will be released this month. Hugging Face CEO Clément Delangue reposted the news.

    Matched fieldsSummary

    #editorial#English text is being prepared.#Model releases#English text is being prepared.
  15. Cailian Press

    Zhipu's Hong Kong shares surged 6.15%, as GLM-5.3 gains recognition with Cursor launch

    Zhipu's Hong Kong-listed shares closed today at HK$665 per share, up 6.15%, with a latest market capitalization of HK$324.2 billion. Cursor announced the launch of GLM-5.3 and GLM-5.3-Flash, stating that GLM-5.3 achieved the highest score of 42.6% among open-weight models on CursorBench 4.0 under the Max reasoning setting; Germany's sovereign AI Aleph Alpha also disclosed that its new model Kolibri uses GLM.

    Matched fieldsSummary、Body

    #chinese#Industry news#OpenAI

Monday

  2. Data Center Knowledge

    Kansas City 20-Story Vertical Data Center Faces Cost and Site Trade-offs

    The 20-story data center tower designed by Chicago architecture firm SOM for developer Revitalization Unlimited is planned for downtown Kansas City, on a site of about 0.23 acres, with 24 MW of IT load and a total development cost of about $183 million, or roughly $1,300 per square foot, higher than suburban facilities. The project team says inference workloads and dense carrier interconnection make the downtown site suitable, but people at IDC and JLL point out that vertical construction still needs to be validated in terms of equipment deployment, cooling water use and overall cost.

    The 20-story data center tower designed by Chicago architecture firm SOM for developer Revitalization Unlimited is planned for downtown Kansas City, on a site of about 0.23 acres, with 24 MW of IT load and a total development cost of about $183 million, or roughly $1,300 per square foot, higher than suburban facilities. The project team says inference workloads and dense carrier interconnection make the downtown site suitable, but people at IDC and JLL point out that vertical construction still

    Matched fieldsSummary、Body

    #hot-signal#Industry news#English text is being prepared.
  3. MarkTechPost

    The Evolution of Qwen: From 7B to 2.4T, a Full Retrospective of Alibaba's AI Model Series

    Alibaba Cloud's Qwen series began with Tongyi Qianwen in April 2023, open-sourced Qwen-7B in August 2023, launched the first Qwen3-Max-Preview with over 1 trillion parameters in September 2025, and ultimately evolved to an open-weight model with 2.4 trillion parameters. In 2024, three generations—Qwen1.5, Qwen2, and Qwen2.5—were released in succession; Qwen2.5's pre-training data grew from 7 trillion to 18 trillion tokens, and model downloads exceeded 40 million. In 2025, Qwen3 adopted a hybrid thinking mode, covering 119 languages, and Qwen3-Next-80B-A3B claims training costs of only 10% of Qwen3-32B and 10x higher inference throughput above 32K tokens.

    Alibaba Cloud's Qwen series began with Tongyi Qianwen in April 2023, open-sourced Qwen-7B in August 2023, launched the first Qwen3-Max-Preview with over 1 trillion parameters in September 2025, and ultimately evolved to an open-weight model with 2.4 trillion parameters. In 2024, three generations—Qwen1.5, Qwen2, and Qwen2.5—were released in succession; Qwen2.5's pre-training data grew from 7 trillion to 18 trillion tokens, and model downloads exceeded 40 million. In 2025, Qwen3 adopted a hybrid

    Matched fieldsSummary、Body

    #ai-vertical#Industry news#Open-source ecosystem#English text is being prepared.