NEWPublish & promote AI tools — get Featured FREEFeatured FREESubmit Tool
← Daily AI researchAI Research

StepFun Releases Step 5 Preview Open Weights: Scale and Efficiency

StepFun Releases Step 5 Preview Open Weights

According to reported information from October 2026, StepFun has released the Step 5 Preview open weights model. Based on the evidence, this model utilizes a 600B Mixture-of-Experts (MoE) architecture, activates 27B parameters, and supports a 1M context window.

Analysis

Analysis of the reported figures suggests a focus on balancing large overall parameter capacity (600B MoE) with optimized active parameter usage (27B active) to manage inference efficiency. Furthermore, the 1M context length aligns with current industry trends toward extended context processing. As per research guidelines, no hands-on testing or benchmark verification has been performed.

Research sources