Qwen3 235B A22B Thinking 2507
Qwen · released 2025-07-25 · apache-2.0 license
Qwen3-235B-A22B-Thinking-2507 is Alibaba's reasoning-mode 235B (22B active) open-weight MoE (July 2025, Apache-2.0), rated the leading open-weights model on the Artificial Analysis index (30): AIME25 92.3, GPQA 81.1, LiveCodeBench 74.1.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 235.09B total · MoE, — active |
| Architecture | qwen3_moe |
| Context window | 262K tokens |
| Knowledge cutoff | 2025-06-30 |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 81.1% |
|---|---|
| SWE-bench Verified | — |
| AIME | 92.3% |
| MMLU-Pro | 84.4% |
| BFCL v3 (tool use) | — |
| Composite score | 7.16 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 137.9 GB | 136.4 GB | — | 262K |
Strengths & weaknesses
Strengths: Leading open-weight reasoning model on the AA index (30) at release; Excellent competition math (AIME25 92.3); Strong coding and 256K long context (RULER 95.4)
Weaknesses: High token usage from long thinking traces; Trails top proprietary on HLE (18.2)