Qwen-3
Third-generation Qwen model featuring hybrid reasoning capabilities that can switch between thinking and non-thinking modes. Trained on 36 trillion tokens (double that of Qwen2.5), with support for 119 languages and dialects. Available in 6 dense models (0.6B to 32B parameters) and 2 MoE models (30B/3B active and 235B/22B active).
2025-04-29
0.6B–32B dense; 30B-A3B and 235B-A22B MoE
Mixture of Experts
Apache 2.0
Specifications
- Parameters
- 0.6B–32B dense; 30B-A3B and 235B-A22B MoE
- Architecture
- Mixture of Experts
- License
- Apache 2.0
- Type
- text
- Modalities
- text
Benchmark Scores
Advanced Specifications
- Model Family
- Qwen
- API Access
- Available
- Chat Interface
- Available
- Multilingual Support
- Yes
- Variants
- Qwen3 dense sizes: 0.6B, 1.7B, 4B, 8B, 14B, 32BQwen3 MoE sizes: 30B-A3B, 235B-A22BInstruct-2507 and Thinking-2507 updates for supported sizes
- Hardware Support
- CUDATPU
Capabilities & Limitations
- Capabilities
- hybrid reasoningthinking modenon-thinking modemathematicscodinglogical reasoningfunction-callingtool usecreative writing
- Known Limitations
- Native context lengths and thinking modes differ across sizes and dated updates.Quantization and serving frameworks are distributions, not independent model identities.
- Notable Use Cases
- agent-based applicationscomplex reasoning tasksmultilingual applicationsAlibaba's Quark AI assistant
- Function Calling Support
- Yes
- Tool Use Support
- Yes
