StepFun Posts Step 5 Preview Weights on Hugging Face Under Its Own Licence
StepFun has published weights for Step 5 Preview, which its model card calls its 'flagship model for agentic work': a 600B-parameter mixture-of-experts with 27B active and up to 1M tokens of context. The StepFun Model License 1.0 is 'Apache-2.0-Style' with extra attribution and commercial-licence conditions. Benchmarks are StepFun's own.
The weights are downloadable, but large commercial users face conditions plain Apache 2.0 would not impose, TSN's reading of the licence shows. StepFun has published weights for Step 5 Preview on Hugging Face; the model card was last updated on October 15, TSN reports. StepFun had already offered the model through its API.
What the card describes
StepFun calls Step 5 Preview its "flagship model for agentic work." The card describes a sparse mixture-of-experts model with 600 billion total and 27 billion active parameters, up to 1 million tokens of context and up to 1 million output tokens. Its specification table lists text and image input and text output.
The card gives deployment paths through vLLM, SGLang and Transformers, with FP8 example configurations on one four-GPU B300 node or one eight-GPU H200 node, and BF16 on one eight-GPU B300 node or two H200 nodes. Benchmark charts on the card are StepFun's own; only the GDPval-AA scores are tied to Artificial Analysis results as of September 20.
The licence
The weights are under the StepFun Model License 1.0, which the card calls "an Apache-2.0-Style license, with additional attribution and commercial-license conditions." According to TSN's reading of the licence text, products above 100 million monthly active users or US$20 million in revenue must display "Powered by StepFun AI," and hardware makers and larger model-as-a-service businesses need a separate commercial licence.
Limits
- The licence thresholds come from TSN's reading; the LICENSE.txt file itself was not opened for this article.
- The card's introduction mentions video input but its specification table does not; this article uses the table.
- No technical report was found, and the benchmark results have not been independently tested.