Qwen 3
Last updated: April 22, 2026
Specs
| Vendor | Alibaba |
|---|
| Released | April 29, 2025 |
|---|
| Context window | 256K tokens |
|---|
| Input | $0.4 / M tokens |
|---|
| Output | $1.6 / M tokens |
|---|
Benchmarks
| mmlu | 86.3 |
|---|
| gpqa | 56.8 |
|---|
| humanEval | 83.5 |
|---|
| arenaElo | 1328 |
|---|
| sweBench | 40 |
|---|
| sweBenchPro | 30 |
|---|
| liveCodeBench | 60 |
|---|
| aiderPolyglot | 52 |
|---|
| terminalBench | 24 |
|---|
| mmluPro | 76 |
|---|
| hle | 7 |
|---|
| aime | 70 |
|---|
| math500 | 89 |
|---|
| mmmu | 65 |
|---|
Strengths
- Strongest Chinese-language performance of any open model
- Open weights with permissive licensing
- 119 languages supported out of the box
Weaknesses
- Less visible in Western benchmarks
- Support ecosystem skewed to APAC
- Lower mindshare among EU/US engineers
Best for
- APAC-facing products needing top Chinese quality
- Multilingual AI with 100+ languages
- On-premise deployments in Asia
Compare with other AI models