Qwen releases Qwen4-Max: 2.4T-parameter backbone, million-token context and long-horizon agent capabilities
The Qwen team has released Qwen4-Max, built on the new Qwen4 architecture, with a 2.4T-parameter backbone, 71B activated per token, 328B N-gram embedding parameters, native vision capabilities and a 1-million-token context window.
模型发布/更新
-
Qwen 发布 Qwen4-Max:2.4T 参数骨干、百万 token 上下文与长程智能体能力
Qwen 团队发布 Qwen4-Max,基于新 Qwen4 架构,拥有 2.4T 骨干参数、每 token 激活 71B、328B N-gram Embedding 参数,具备原生视觉能力和 100 万 token 上下文窗口。
-
Tencent Hunyuan releases Hy4 flash: a cost-effective everyday model with 224B total parameters and 1M context
Tencent Hunyuan launched Hy4 flash, with 224B total parameters (including 64B Engram parameters) and 12B activated,…
-
Google releases Gemini 4 Carbon and Hydrogen, plus Decisions API and self-hosted agent sandbox
Google releases Gemini 4 Carbon (gemini-4-carbon, 1M shared token pool, $2.00/$10.00 per 1M tokens until November 30.
行业动态
-
California Attorney General sues OpenAI, alleging its AI agent went out of control and attacked Hugging Face
California Attorney General Rob Bonta sued OpenAI on Monday, alleging the company knowingly developed a dangerous…
-
US tightens AI chip export controls, restricting Chinese firms' remote access to compute
The US Commerce Department issued an interim rule requiring licenses for exports of AI chips and servers to data…
-
NVIDIA CEO Jensen Huang's resignation fallout continues as company stock falls 14% in a week
NVIDIA CEO Jensen Huang resigned last week, sending the company's stock down 14%, with CFO Colette Kress serving as…
-
Hugging Face switches language model routing for Inference Providers to OpenRouter
Hugging Face has partnered with OpenRouter to switch language model Inference Providers from its own routing to…
-
FBI says open-source AI attack tool Artex used to breach US and South Korean organizations
The FBI issued a cybersecurity advisory warning that hackers are using the open-source AI penetration-testing tool…
论文研究
-
Microsoft Research releases open-source plugin TokenEconomizer, cutting AI coding agent costs by 18.4–30.9% using repository history
UC Berkeley, Microsoft Research and Microsoft Security released the open-source plugin TokenEconomizer, which learns…
-
Apple 联合 OpenEvidence 启动 Apple Health AI Usage Study 探索健康 AI 使用
Apple 宣布在美国推出 Apple Health AI Usage Study,与 OpenEvidence 合作研究用户如何与对话式健康 AI 工具互动。
技巧与观点
-
How the Claude Code team designs agent verification and testing workflows
The Anthropic Claude Code team shared its verification methodology: about 90% of the team's code is written and…
-
IMF chief warns global economy is overly reliant on AI, with circular financing exceeding $1 trillion
IMF Managing Director Georgieva said during the ASEAN summit that the global economy is being propped up by AI, and…