← Back to blog
Step 5 Preview: Features, Pricing, Benchmarks & Access

Step 5 Preview: Features, Pricing, Benchmarks & Access

Alex Morgan

By Alex Morgan

MyClaw Editorial

MyClaw

Run Best-in-Class AI Agents Now

Run OpenClaw or Hermes Agent on managed MyClaw hosting, so your AI agent stays online, updated, and ready for real work.

AI Takeaway

  • What is Step 5 Preview?It is StepFun’s model for coding, knowledge work, and agent tasks, with a 1M-token context window and text, image, and video input.
  • How much does it cost?Official API rates are $1 per million uncached input tokens, $0.05 for cached input, and $2.70 for output, including reasoning tokens.
  • Can you access it now?Yes. StepFun documents step-5-preview for API use, subject to account permissions. Standard API usage and Step Plan have separate billing arrangements.
  • When should you choose it over Step 3.7 Flash?Consider it when you need more than 256K context or it performs better on your task. Step 3.7 Flash has lower token prices and also accepts images and video.
  • Can you use it in a MyClaw agent?That depends on provider and runtime compatibility. MyClaw supplies managed OpenClaw hosting; access to this specific model needs separate verification.

What Is Step 5 Preview?

Step 5 Preview is a StepFun reasoning model aimed at programming and professional knowledge work. Its official documentation and pricing are available as of September 20, 2026, superseding earlier coverage that described it as unannounced.

Here is what StepFun currently documents:

FeatureStep 5 Preview
API model IDstep-5-preview
Context window1 million tokens
InputText, images, video
OutputText
Tool callingSupported
Structured outputJSON Mode and JSON Schema
Reasoning effortLow, medium, high

The practical question is whether these capabilities help with your workload: a larger document collection, a difficult repository change, or a task combining written instructions with visual evidence. These are useful starting points for evaluation.

The surrounding application supplies browsing, file access, and code execution. The model requests actions and interprets the results. This guide to how OpenClaw connects models and tools explains that division of work.

Step 5 Preview Benchmarks and Practical Uses

As checked on September 20, 2026, Artificial Analysis reported an Intelligence Index score of 44 and output throughput of 99.8 tokens per second for Step 5 Preview. This article draws on published evaluations and documentation; it does not report original hands-on testing.

The index is a benchmark score, not a percentage of tasks the model will complete correctly. Throughput measures generation speed; input processing, reasoning, tool execution, and retries also affect how long a finished task takes.

To compare models fairly, keep the sources, tools, instructions, and success criteria consistent. Repeat the assignment and record failed attempts as well as good results.

StepFun on X: "Introducing Step 5 Preview: Advancing the Pareto Frontier. Step  5 Preview is our new flagship model for agentic work, delivering  frontier-level performance across software engineering and professional  knowledge work,

Coding and Repository Work

For a repository task, combine an error log, related source files, and the expected behavior. Ask for a minimal fix and a regression test. This exercises the model’s ability to connect evidence across files and follow a change through verification.

In ongoing coding agent workflows, assess patch scope, test results, and the manual corrections required. Count a task as complete when the change works and can be reviewed, rather than when the model produces an explanation.

Long Documents, Images, and Video

A research assignment could compare product documentation with a chart and a short demonstration video, then identify where the sources disagree. Ask for document references and video timestamps so the conclusions are easy to check.

In research agent workflows, separate supported findings from inference and unresolved questions. More context allows more material in a request, but source selection still matters. Remove duplicate pages and label versions clearly; an outdated document should not silently override its replacement.

Step 5 Preview Pricing vs. Step 3.7 Flash

Step 3.7 Flash also supports text, images, and video, making it a useful lower-cost comparison. Its context window is 256K. Prices below are StepFun’s published USD rates per million tokens, checked September 20, 2026.

DetailStep 5 PreviewStep 3.7 Flash
Context window1M256K
Uncached input$1.00$0.20
Cached input$0.05$0.04
Output$2.70$1.15

With 100,000 uncached input tokens and 10,000 billed output tokens, the hypothetical costs are:

  • Step 5 Preview: $0.127 — $0.10 input plus $0.027 output.
  • Step 3.7 Flash: $0.0315 — $0.02 input plus $0.0115 output.

Step 5 Preview: China's Cheapest Frontier AI Model Yet

That comparison holds token counts constant and excludes other services. Actual tasks may consume different amounts with each model.

Step 5 Preview bills reasoning tokens as output alongside the final answer. Estimate from recorded usage, including confirmed cache hits, rather than answer length alone.

Start with Step 3.7 Flash for work that fits its context and meets your quality requirements. Try Step 5 Preview when you need more room or when a better result could justify the premium. Image or video input alone is not a reason to upgrade.

Budget separately for model usage and managed hosting costs. Step Plan has its own allowances; neither those benefits nor a hosting subscription should be assumed to cover standard API charges.

Skip the Setup, Run Best-in-Class AI Agents Now

Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.

Get Started

How to Access Step 5 Preview

Standard API or Step Plan

For standard API access, create a StepFun Open Platform key and confirm your account can use step-5-preview. Configure the platform’s API base URL in an OpenAI-compatible client, select that exact model ID, and send a short Chat Completions request. Check the response and recorded usage before adding tools.

For Step Plan, use the integration guide for your client. StepFun’s Claude Code guide specifies a subscription endpoint containing /step_plan; removing that path switches to the standard API channel. Confirm the intended billing channel in the console.

The Step Plan page checked September 20 advertises 15 days of access, with another 15 after a first API call. Eligibility and claim terms apply; this is a temporary promotion.

Check the client’s context display, too. StepFun documents that Claude Code may default to 200K for an unrecognized model. Its guide describes enabling the 1M window and checking it with /context after restarting.

Check the Tools Your Agent Actually Uses

Next, try a harmless tool task: read a test document and save a summary. Confirm that the file exists and its contents match the source.

For web work, connect browser automation tools and inspect the retrieved page. This separates a working model connection from a working tool integration before you depend on either for a longer assignment.

Run a Repeatable Research Workflow with MyClaw

Once a research task becomes recurring, it needs a persistent workspace, source access, and a schedule. MyClaw provides managed hosting for a private, always-on OpenClaw instance so the workflow can continue while your computer is offline.

Step 1: Set Up Your Research Workspace

Choose managed OpenClaw hosting and define one narrow assignment. For example: track changes to a product’s documentation and produce a brief explaining what affects your integration.

List the exact pages and save a dated starting snapshot. Specify the changes that matter: API parameters, prices, limits, or deprecations. That gives the agent a concrete comparison instead of a vague request to summarize whatever looks new.

Get Started

Step 2: Verify the Model Connection

Step 5 Preview availability in MyClaw remains unverified for this article. Check that your instance supports the required provider configuration and that your account can call the model. If both are confirmed, run the documentation task through that connection.

Otherwise, begin with an already supported model. Keep the source snapshots and instructions so you can evaluate Step 5 later without rebuilding the workflow.

Step 3: Test the Report, Then Schedule It

Connect the required tools and run the report manually first. Ask for the previous wording, current wording, source location, and practical impact of each change. Spot-check those entries against the saved pages.

Once that output is dependable, schedule repeat runs. Have the agent distinguish “no changes found” from “page could not be checked,” and record completion time and usage cost. Review the first few reports before expanding the source list.

Should You Try Step 5 Preview?

Step 3.7 Flash is a sensible default when it meets your quality and timing needs within its context limit. Try Step 5 Preview when that limit becomes restrictive or difficult assignments still require substantial correction.

Compare completion time, total cost, and corrections on one representative job. Choose the model that earns its place in that workflow, then schedule recurring work once the connection and report quality are reliable.

Skip the Setup, Run Best-in-Class AI Agents Now

Launch a managed OpenClaw or Hermes Agent workspace in minutes, with always-on hosting, updates, and support handled by MyClaw.

Step 5 Preview: Features, Pricing, Benchmarks & Access | MyClaw.ai