Qwen3-Next
Announced on qwen.ai
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| AIME 2025No tools · thinking model | 87.8% | |
| instruct model | 69.5% | |
| MMLU-Prothinking model | 82.7% | |
| HMMT February 2025No tools · thinking model | 73.9% | |
| LiveCodeBenchthinking model; LiveCodeBench v6, problems from Feb to May 2025 | 68.7% | |
| instruct model; LiveCodeBench v6, problems from Feb to May 2025 | 56.6% | |
| τ²-benchTelecom · thinking model | 43.9% | |
| τ²-benchRetail · thinking model | 67.8% | |
| τ²-benchAirline · thinking model | 60.5% |
Sources
Reported by Alibaba
- Qwen/Qwen3-Next-80B-A3B-Thinking model card (launch-day revision)huggingface.co · 11 Sep 2025
- Qwen3-Next: Towards Ultimate Training & Inference Efficiencyqwen.ai · 11 Sep 2025
Original image 
Original image
