Stepfun Open-Sources Step 3.7 Flash LLM Optimized for Agent Era

Stepfun has open-sourced Step 3.7 Flash, a sparse MoE language model with 196B total parameters and 11B active parameters.

The model is built for agent-era workloads, including multimodal understanding, web and visual search, coding, and tool orchestration.

Step 3.7 Flash is designed for low-latency use, with reported output speeds of up to 400 tokens per second.

It is compatible with major agent frameworks and tool-calling protocols, including MCP and several popular coding agents.