Step 5 Preview is StepFun’s flagship model for agentic work, with stated strengths in software engineering, professional knowledge work, and finance.
It natively accepts text, image, and video input and can use tools and documents for multi-step tasks.
| Developer | StepFun |
|---|---|
| Model ID | stepfun/step-5-preview |
| Release Date | October 8, 2026 |
| Context Window | 1,000,000 tokens (~1M) |
| Max Output Tokens | 64,000 tokens |
| Modalities | Image, Text, Video → Text |
| Capabilities | reasoning, streaming, structured (JSON schema) output, tool/function calling, vision |
Agentic work
StepFun describes Step 5 Preview as a flagship model for agentic work and says it can support tasks involving tool calls and continued progress toward a deliverable.
Multimodal input
The model natively supports text, image, and video input.
Reasoning controls
Step 5 Preview supports low, medium, and high reasoning effort levels.
Structured responses
The model supports JSON Mode and JSON Schema structured outputs.
Tool integration
Tool calling is supported for the model.
Software engineering
StepFun identifies software engineering as one of the model’s areas of strength.
Professional knowledge work
The model is positioned for professional knowledge work, including tasks that require handling information and progressing toward a deliverable.
Finance
StepFun states that Step 5 Preview has particular strength in finance.
Multimodal analysis
The model can be used for analyzing images, charts, screenshots, and video content.
Sources: https://platform.stepfun.ai/docs/en/guides/models/overview · https://platform.stepfun.ai/docs/en/guides/models/step-5-preview · https://platform.stepfun.com/docs/zh/guides/models/step-5-preview
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Software development: locating problems from code, dependency information, and error logs; modifying code; and recommending tests.
Long-document processing: organizing multiple sources, extracting and summarizing information, and cross-checking details.
Research and analysis: using search, code-execution, or document tools to break down tasks and produce structured results.
Multimodal understanding: analyzing screenshots, charts, and videos, then presenting conclusions in a specified format.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "stepfun/step-5-preview", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "stepfun/step-5-preview", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"stepfun/step-5-preview","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
Cache-miss input billing includes writing new content to the cache, output billing includes both reasoning tokens and the final answer, and API requests are rate-limited by account tier based on cumulative top-up amount.
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 43.7 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | October 8, 2026 |
The primary published source is StepFun’s official Step 5 Preview announcement, “Step 5 Preview: Advancing the Pareto Frontier.” Source: https://www.stepfun.com/step-5-preview
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Step 5 Preview This page | Agentic workflows and structured output | |||
| Coding + agents |
Step 5 Preview is StepFun’s flagship model for agentic work, with strengths in software engineering and professional knowledge work, particularly finance.
Step 5 Preview natively supports text, image, and video input.
Yes. Step 5 Preview can understand images and videos without requiring an additional vision model.
Yes. Step 5 Preview supports low, medium, and high reasoning effort levels.
Yes. Tool calling is supported for Step 5 Preview.
Yes. Step 5 Preview supports both JSON Mode and JSON Schema structured outputs.
Yes. Streaming output is supported for Step 5 Preview.
It can combine images, video, and text to extract and analyze multimodal information.
Yes. It is intended for tasks involving large amounts of information, tool calls, and continuous progress toward a deliverable.
Step 5 Preview uses a sparse Mixture-of-Experts architecture with 600 billion total parameters and 27 billion active parameters per token.
