StepFun 3.7 Flash Released: 196B Multimodal MoE Model Runs Locally

Core summary

One sentence to understand this update

StepFun released its 3.7 Flash model, a 196B total / 11B active multimodal Mixture-of-Experts (MoE) model that runs locally on 128GB RAM, featuring a built-in 1.8B ViT for vision.

Impact & opportunity

What this could mean

Developers looking to deploy large multimodal models on local devices can consider StepFun 3.7 Flash for high-performance, offline AI capabilities.