About this model or application
Step Flash models use sparse MoE inference and configurable reasoning. Step 3.7 adds a vision encoder; parameter accounting distinguishes backbone and auxiliary modules.
Family & other versions2 sources
Background for the family or other versions. These sources are counted separately from evidence for this release.
Model cards & docsOfficial
Step-3.7-Flash official documentationFamily & other versions
Official model specifications and deployment information.
Model cards & docsOfficial
Step-3.5-Flash official documentationFamily & other versions
Official model specifications and deployment information.
Model family catalog2
Current models
Official catalogStep-3.7-FlashAvailable256K ContextOpen weights
- Model ID
stepfun-ai/Step-3.7-Flash- Released
- Not stated
- Availability
- Available
- Input
- Text · Image
- Output
- Text
- Context
- 256,000 tokens
- Parameters
- 198B / 11B active
- Deployment / access
- Open weights · Local
- Weights
- Open weights
- License
- Apache-2.0
- Capabilities
- Reasoning · Vision
198B total and 11B active are rounded figures; language backbone 196B and vision encoder 1.8B.
Step-3.5-FlashAvailable256K ContextOpen weights
- Model ID
stepfun-ai/Step-3.5-Flash- Released
- Not stated
- Availability
- Available
- Input
- Text
- Output
- Text
- Context
- 256,000 tokens
- Parameters
- 196.81B / 11B active
- Deployment / access
- Open weights · Local
- Weights
- Open weights
- License
- Apache-2.0
- Capabilities
- Reasoning · Tool use
Total includes 0.81B auxiliary head; activated count is approximate.