Qwen Chat
Alibaba’s Qwen models in a free, capable chat interface.
Latest Qwen Chat News
EU Market Opens, Compliance Becomes the Gate
As the EU AI Act enters mandatory enforcement, Chinese model providers face transparency, copyright, and systemic-risk obligations when entering the European market. European companies are nevertheless adopting DeepSeek, Qwen, Kimi, and GLM because of lower costs, local deployment options, and concerns about dependence on U.S. suppliers.
Alibaba releases Qwen3.8-Max ahead of open-weights launch
Alibaba Group has made its flagship Qwen3.8-Max model widely accessible to global users. This release precedes a full open-weights launch scheduled for next week, signaling a shift back to open-source strategies for their top-tier models.
llama.cpp GPU Sampling Delivers Faster Decoding
A llama.cpp pull request moves sampling from the CPU to the GPU for users running multi-token prediction (MTP). The contributor reports roughly an 8% improvement on an RTX 5090 and about 4% on a Tesla P40 with Qwen3.6-35B, while maintaining the same acceptance ratio.
Qwen Office Gives Agents Task Authority
Alibaba's Qwen Office introduces an enterprise-facing office Agent with authority to allocate tasks. The update signals a shift from AI assistants executing requests to Agents coordinating organizational workflows.
Qwen Code v0.21.5 Strengthens Agent Reliability
Qwen Code v0.21.5 adds safer tool-call recovery, execution outcome tracking, and protections against infinite Goal mode retries. It also introduces a macOS migration bridge to the Tauri shell, structured Web Shell review findings, and Qwen 3.8 reasoning-effort support.
MiniMax H3 Goes Open Source, Qwen3.8-Max Arrives
少數派的 AI 近期動態整理提到 MiniMax H3 開源，以及 Qwen3.8-Max 發布。這兩項模型消息對希望測試開源或新一代大語言模型的開發者值得關注。