Qwen3-Coder-480B-A35B and Qwen Code
480B-A35B open agentic coding model, 256K native context (1M with YaRN), claimed best open model on agentic coding and competitive with Claude Sonnet 4.
Pretrained on 7.5T tokens (70% code), post-trained with execution-driven RL and long-horizon Agent RL across 20,000 parallel environments. Shipped with Qwen Code, a CLI adapted from Gemini CLI, and Claude Code and Cline compatibility.
- Date
- Tuesday, 22 July 2025
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Pretraining tokens | 7.5T (70% code) Context 256K native, 1M with YaRN; Agent RL on 20,000 parallel environments; SOTA among open models on SWE-bench Verified per Alibaba | company |
SWE-bench claims are Alibaba's. License not stated on the blog page I opened (the Hugging Face cards of the family use Apache 2.0).
Sources
This record was checked against its sources on 6 October 2026. How we check