Step
Step 3.5 Flash — a 196B sparse MoE with about 11B parameters active per token.
Step 3.5 Flash is the largest sparse MoE in the matrix by total parameter count. It keeps roughly 11B parameters active per token, so decode cost tracks the active set rather than the 196B total. It was added in v0.5.3 and uses the same OGA + HIP EP path as the other text models.
Step 3.5 Flash
Added in v0.5.3. A 196B MoE that activates about 11B parameters per token.
- Kind: Language model
- Task: Text generation, reasoning
- Parameters: 196B total, 11B active
- Architecture: Sparse mixture of experts
- Quantization: int4 RTN, group 128
- Base model: stepfun-ai/Step-3.5-Flash
- Validation status: New in v0.5.3
- Measured in v0.5.3: 16.99 tokens/s, 30.09 s to first token, at a 2048-token prompt — all three lengths