StepFun has made its Step 5 preview model available as open weights. The checkpoints, in FP8 and BF16 formats, are posted on Hugging Face and ModelScope. The model is a sparse Mixture‑of‑Experts (MoE) system with 600 billion total parameters but only 27 billion active per token. It can handle contexts up to 1 million tokens, aimed at long‑horizon agentic tasks.
Why it matters
Developers can now experiment with a very large, open‑weight model that supports much longer text inputs than typical LLMs.
