StepFun opens Step 5 Preview weights: 600B MoE agent model with 1M context
StepFun released open weights for Step 5 Preview, a 600B-total / 27B-active sparse MoE pitched at repo-level software engineering and long-horizon tool use, with up to 1M tokens of context and output. FP8 runs on one 4x B300 or 8x H200 node via day-0 vLLM and SGLang images, under the Apache-style StepFun Model License 1.0 with extra commercial terms.