Details
- Qwen unveils Qwen3.8-Max, described as its most capable model so far, with 2.4 trillion parameters.
- The company states that Qwen3.8-Max sets a new bar for coding and cowork use cases, emphasizing autonomous development workflows.
- Qwen announces that open weights for Qwen3.8-Max will be released next week, expanding access for external developers and researchers.
- Qwen3.8-27B is also going open-weights, offering a smaller but still advanced alternative model for community use.
- The thread highlights long-horizon autonomous coding capabilities, including experiments running 10 to 16 days of continuous coding.
- Qwen links language model (LM) performance dashboards and separate English-language performance reports to showcase benchmark results.
- Visual-language (VL) performance resources and demos are shared, indicating strong multimodal understanding and visual agentic intelligence.
- The announcement emphasizes cowork and dynamic workflows, positioning Qwen3.8-Max as a general AI collaborator that can take on any role in complex projects.
- Qwen references quant strategy workflows and Text Arena evaluations, suggesting competitive standing against other frontier models.
- Multiple linked resources point to technical reports and evaluations that detail Qwen3.8 series performance across coding, reasoning, vision, and agentic tasks.
Impact
Qwen’s decision to release open weights for both a 2.4T-parameter flagship (Qwen3.8-Max) and a 27B variant strengthens the open frontier-model ecosystem and pressures closed competitors that dominate coding and cowork agents. By combining autonomous long-horizon coding, multimodal agentic behavior, and accessible weights, Qwen positions itself closer to leading players in AI development platforms and may accelerate research on large-scale agentic coding and quant workflows.
