Qwen3 Next 80B A3B Thinking
Qwen · released 2025-09-11 · apache-2.0 license
Qwen3-Next-80B-A3B-Thinking is Alibaba's efficiency-focused reasoning MoE (80B total, 3B active, Sept 2025, Apache-2.0): AIME25 87.8, GPQA 77.2, MMLU-Pro 82.7, LiveCodeBench 68.7. Artificial Analysis index 27.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 81.32B total · MoE, — active |
| Architecture | qwen3_next |
| Context window | 262K tokens |
| Knowledge cutoff | 2025-09-30 |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 77.2% |
|---|---|
| SWE-bench Verified | — |
| AIME | 87.8% |
| MMLU-Pro | 82.7% |
| BFCL v3 (tool use) | — |
| Composite score | 8.25 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 48.7 GB | 47.2 GB | — | 262K |
Strengths & weaknesses
Strengths: Efficient reasoning MoE (80B total, 3B active) Qwen reports beating Gemini-2.5-Flash-Thinking; Strong math/coding at tiny active count (AIME25 87.8); AA index 27, above average for its class
Weaknesses: Verbose reasoning; Trails the larger 235B Thinking on the hardest tasks