StepFun

Step 5 Preview

StepFun's flagship preview for coding, research and professional analysis. A sparse 600B-parameter model activates 27B parameters per token.

Preview releaseText + visiontext + image inputtext output

Run this model

stepfun/step-5-preview

Use an endpoint supported by this model.

Get a key for this setup →
const response = await fetch('https://api.minirouter.sh/v1/chat/completions', {  method: 'POST',  headers: {    Authorization: `Bearer ${process.env.MINIROUTER_KEY}`,    'Content-Type': 'application/json',  },  body: JSON.stringify({    model: 'stepfun/step-5-preview',    messages: [{ role: 'user', content: 'Why is the sky blue?' }],    max_tokens: 1024,  }),}) const data = await response.json()console.log(data.choices[0].message.content)
Read docs →

Use stepfun/step-5-preview with Chat Completions and an explicit max_tokens limit. Vercel currently lists max_tokens, temperature and stop; do not assume tools, tool_choice or configurable reasoning are supported. Generic client setup examples do not establish agent compatibility.

Model overview

Input / 1M
$1.05
All-in MiniRouter rate
Output / 1M
$2.835
All-in MiniRouter rate
Context
1M
1,000,000 tokens
Max output
1M
tokens
Architecture
600B MoE
27B active per token
Open weights
Planned Oct 15
Publisher commitment, not released yet
Announced
20 Sept 2026
publisher release

Publisher evidence

Evaluate coding and professional work separately.

StepFun reports the following high-effort results. DeepSWE uses SWE-agent with temperature 1.0 and top_p 0.95. Each chart is a separate test; the percentages do not form a composite score or a MiniRouter performance claim.

StepFun-reported

DeepSWE v1.1

High-effort evaluation. Use the publisher's methodology when comparing with another model.

Step 5 Preview · high

0 to 100%

  • Step 5 Preview67.7%
StepFun Step 5 Preview announcement ↗, checked September 21, 2026.
View chart data
DeepSWE v1.1 value
Model / effortReported result
Step 5 Preview / high67.7%

StepFun-reported

Terminal-Bench 4.0

High-effort evaluation. Use the publisher's methodology when comparing with another model.

Step 5 Preview · high

0 to 100%

  • Step 5 Preview33.3%
StepFun Step 5 Preview announcement ↗, checked September 21, 2026.
View chart data
Terminal-Bench 4.0 value
Model / effortReported result
Step 5 Preview / high33.3%

StepFun-reported

FrontierFinance

High-effort evaluation. Use the publisher's methodology when comparing with another model.

Step 5 Preview · high

0 to 100%

  • Step 5 Preview66.4%
StepFun Step 5 Preview announcement ↗, checked September 21, 2026.
View chart data
FrontierFinance value
Model / effortReported result
Step 5 Preview / high66.4%

Trying the preview

Bring long source material and clear acceptance criteria.

The model's context and vision support make document-heavy tasks worth evaluating. Check the supported parameters before connecting an agent.

See the release and independent measurements
  1. 01

    Code and research

    Try bounded code changes, document synthesis and evidence reconciliation. Keep tool execution in your application until the catalog advertises the tool parameters it needs.

  2. 02

    Visual source material

    Provide an image alongside a text question. The endpoint returns text; visual input does not imply image, video or audio generation.

  3. 03

    Analytical reports

    Verify calculations and source citations before relying on an analysis. Compare the quality and total tokens of a completed task, not just the output rate.

Costs and preview status

A context ceiling is not an output budget.

Vercel advertises 1M context and a 1M output ceiling. Input and generated output share that window; choose a smaller explicit output limit for your task.

Current provider listing
  1. 01

    Use the catalog price

    MiniRouter's current prices appear above. Provider list prices and cache-read rates are documented separately in the release article; caching is not guaranteed for every request.

  2. 02

    Keep the preview name

    Select stepfun/step-5-preview explicitly. It is not an alias for Step 3.7 Flash, and an existing Flash configuration may send unsupported tool fields.

  3. 03

    Open weights come later

    StepFun plans an October 15 weight release. API access today does not establish a downloadable model, a license or self-hosting requirements.

One route, standard clients.

Authenticate with a MiniRouter bearer key. Caller-provided upstream credentials are not used on this route.

Read model setup docs →
Base URLhttps://api.minirouter.sh/v1
Model IDstepfun/step-5-preview
Endpoints
/v1/chat/completions/v1/messages/v1/responses
Parameters

max_tokens, temperature, stop, reasoning, include_reasoning

Structured output

Not enforced; a requested format is dropped and the answer is plain text.

Configured provider

Priority follows the live catalog order.

Input: text, image · Output: text

ProviderPriorityInput / 1MOutput / 1MParametersAccess
vercel
1$1.05$2.835max_tokens, temperature, stop, reasoning, include_reasoningMiniRouter balance