Key Info

StepFun's Step 5 Preview is now available on GMI Cloud. It is a 600B-parameter sparse MoE model that activates 27B parameters per token, with a 1M-token context window and native text and image understanding.

Highlights

  • Sparse MoE architecture: 600B total parameters, 27B active per token
  • 1M-token context window for long-horizon tasks
  • Native text and vision input support
  • Targeted at agentic work in coding, finance, and professional workflows
  • Open weights released in BF16 and FP8 on Hugging Face and ModelScope, alongside a technical blog