Alibaba's Qwen team open-sourced Qwen3-Coder-480B-A35B-Instruct, an open agentic coding model with 256k native context (extensible to 1M) and top-tier SWE-bench Verified marks. Alongside the weights, Qwen open-sourced Qwen Code, a terminal agent CLI for autonomous project refactoring.
Key Takeaways
- ✓480B total / 35B active MoE architecture sets new standards on SWE-bench Verified.
- ✓256k native context with 1M extrapolation ingests large monolithic codebases.
- ✓Open-source Qwen Code CLI delivers an out-of-the-box autonomous terminal agent.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.