← Back to Model Beat
Models·1d ago·all news from September 21, 2026

StepFun Launches Step 5 Preview: A 600B-Total, 27B-Active MoE Model With 1M Context for Long-Horizon Agentic Work

StepFun has released a preview of its Step 5 model, a sparse Mixture-of-Experts architecture featuring 600 billion total parameters and 27 billion active parameters. The multimodal model supports input for text, images, and video alongside a 1-million-token context window. Designed for long-horizon agentic tasks, it aims to assist users in fields like professional research and software development.

ModelsStep 5 Preview

Covered by 1 source

Related stories

ModelsGoogle Joins OpenAI, Anthropic, Meta in Disclosing AI HacksSep 18 · 21 sourcesModelsSecurity researchers used Anthropic's Claude to hack OpenAI's internal systems in under 72 hoursSep 18 · 9 sourcesModelsAlibaba's open-weight Qwen-Image-2.1 claims to beat closed models in image generation with just 7 billion parametersSep 20 · 12 sourcesModelsQwen3.8-Omni-Flash undercuts Google's Gemini Flash pricing while matching its multimodal benchmarksSep 18 · 10 sources