Qwen3 Coder 32B
- Lab
- qwen
- Weights
- open_weights
- Released
- 2026-06-30
- Context window
- 128,000
- Modalities
- text
- API
- yes
- Local-runnable
- yes
- Status
- current
announcementmodel cardpricing
Scores
Coding
| Value | Source | As of | Reported by | Notes |
| 58.9 percent | swe-bench-verified ↗ | 2026-08-20 | independent | Seed placeholder — best open score in its size class. |
Output speed
| Value | Source | As of | Reported by | Notes |
| 24 tokens_per_second | owner-notes ↗ | 2026-08-10 | independent | Seed placeholder — 4-bit MLX on M-series 64GB, owner-measured. |
Cost
| Value | Source | As of | Reported by | Notes |
| 0.9 usd_per_million_blended | artificial-analysis-pricing ↗ | 2026-08-24 | aggregatorA | Seed placeholder — hosted API price; local is free. |
Runs well locally on 64GB Apple Silicon
| Value | Source | As of | Reported by | Notes |
| yes | owner-notes ↗ | 2026-08-10 | independent | 4-bit quant fits in ~20GB; comfortable on 64GB. |
Signals
- hf_downloads_30d: 96,000 as of 2026-08-25 source
Verdicts
Events
3 Qwen released an MIT-licensed agent runtime with sandboxed tool execution and long-horizon memory, reporting a 6-point Terminal-Bench gain for Qwen3 Coder 32B when run inside it. Curators flag it as the most credible open alternative to proprietary coding agents.
2026-08-05 Tooling / agents Open weights Qwen blog — agent runtime release ← Back to Models