AI BriefingStepFun Unveils 3.7 Flash Model
Key point
StepFun has released Step 3.7 Flash, a 196B-scale MoE-based multimodal model.
Details
StepFun has released Step 3.7 Flash, a new multimodal MoE model. This model has a total of 196B parameters, but through its MoE (Mixture-of-Experts) architecture, it uses only 11B active parameters during inference. It also has a built-in 1.8B ViT to support vision capabilities.
Key benchmark results are as follows:
- SWE-Bench Pro: 56.26% (on par with or exceeding DeepSeek V4 Flash and Gemini 3.5 Flash)
- DeepSearchQA F1: 92.82% (competitive with GPT 5.5)
- HLE (with tools): 47.2%
This model shows very high performance relative to its active parameter count on agent and coding tasks. It is currently available through OpenRouter and NVIDIA NIM.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis PerformancePrivatemode Benchmarks GLM-5.3-Flash as Vision-Capable Decision Model with Parity to Jev but Higher Cost