• Sources: primary, discussion
  • Summary: Step 5 Preview is a sparse MoE with 600B total and 27B active parameters, a context of up to 1M tokens, and text and image input. The model card gives vLLM, SGLang and Transformers recipes, with FP8 on one 4-GPU B300 node or one 8-GPU H200 node and BF16 on two H200 nodes. It ships under a custom license marked other.
  • Why it matters: Teams that self-host get a new open-weight candidate for long-horizon coding agents at a single-node FP8 footprint.