Step 5 Preview
StepFun's Step 5 Preview is a 600B-parameter open-weights model trained for long-running agent tasks.
Why it matters
A new large open-weights model from StepFun aimed at agentic coding and long, tool-heavy knowledge work, with a particular focus on finance. You can use it through StepFun's API or self-host the weights.
Highlights
- Sparse mixture-of-experts: 600B total parameters with 27B active per token.
- Context of up to 1M tokens. The API docs list text, image and video input, text output, and a 64K maximum output; the Hugging Face card lists text and image input.unverified
- The API supports tool calling, JSON Mode and JSON Schema output, low/medium/high reasoning effort, and prompt caching.
- Can be served with vLLM or SGLang through an OpenAI-compatible API, or run with Transformers. FP8 and BF16 checkpoints are available.
- A "narrow but deep" design with 92 transformer blocks, per StepFun's architecture post.unverified
Taken from the project's own README, docs or launch page. Claims about speed or accuracy are the maker's unless stated otherwise.
Worth knowing: Benchmark results on the model page are StepFun's own. The official page renders with JavaScript and could not be read directly; facts come from the Hugging Face card and StepFun docs.
Quick facts
| Pricing | StepFun API is billed per input/output token (rates not checked). Third-party host Featherless lists $1 per 1M input and $2.70 per 1M output tokensnot verified |
|---|---|
| Free or paid | Open weights (free to download); paid API |
| License | StepFun Model License 1.0 (Apache-2.0-style, with extra attribution and commercial-license conditions) |
| Language | Model (any language via API) |
| Platform | StepFun API; self-host via vLLM / SGLang / Transformers |
| GitHub stars | n/a (not a GitHub project) |
| Maker | StepFun |
Links
- Official site: https://stepfun.com/step-5-preview
- Repository: https://huggingface.co/stepfun-ai/Step-5-Preview
- Release notes: https://platform.stepfun.ai/blog/inside-step-5-preview
- Docs: https://platform.stepfun.ai/docs/en/guides/models/step-5-preview
Related ranking
See how established tools compare in Best AI Agents (General & Computer-Use). New releases aren't scored until we review them against the rubric.
What we couldn't verify
- quick_facts.pricing (StepFun's own API rates)
- highlights[1] (input modalities differ between API docs and model card)
- highlights[4] (from search extract of StepFun's blog)
Checked 2026-10-17 from the model's own pages. Prices and star counts change, so confirm on the official site. Listing here is not a ranking or an endorsement.
The week's best new AI tools, ranked
One email a week: new releases worth your time, ranking changes, and one deep-dive. No spam.