StepFun Releases Step 3.7 Flash: Efficient Open-Source MoE VLM
Key Questions
What is Step 3.7 Flash and its key specifications?
Step 3.7 Flash is a 198B total / 11B active parameter MoE vision-language model with 256k context length. It is released under the Apache 2.0 license by StepFun.
How does Step 3.7 Flash perform compared to other models?
It achieves 97% of Opus 4.6 performance at one-ninth the cost in Advisor Mode. The model shows strong results on coding agent and multimodal benchmarks.
What does the release of Step 3.7 Flash indicate for open-source AI?
It signals a new wave of efficient open-source models optimized for agentic workflows. The model targets coding agents and search use cases.
StepFun released Step 3.7 Flash, a 198B total / 11B active MoE vision-language model with 256k context, Apache 2.0 license. Achieves 97% of Opus 4.6 performance at 1/9 cost in Advisor Mode. Strong on coding agent and multimodal benchmarks, signaling a new wave of efficient open-source models for agentic workflows.