Step 5 Preview

What is Step 5 Preview?

Step 5 Preview is StepFun’s model for software and professional knowledge workflows involving several steps. OpenRouter lists text, image and video inputs with text output. It is a candidate for a developer building a research or coding assistant, with the preview status an important reason to validate changes before relying on it.

Give a multi-step task a concrete deliverable

StepFun documents repository-level engineering and analysis of documents, charts and screenshots. A useful trial is a migration note: provide the relevant files, the target API documentation and one failing example, then ask which changes are necessary and which remain uncertain.

Another candidate is a document comparison. Supply two dated reports and request a list of changed assumptions with source locations. Keep interpretation separate from extracted numbers, and calculate important totals independently. The aim is a reviewable working note, not an unqualified business recommendation.

Suggested workflow: map a small migration

Select one dependency change and a limited set of files. Give the model a clear completion condition, such as removing one deprecated call while preserving the existing output. Ask for a file-by-file plan before allowing edits.

Run the proposed change in a test environment and compare the resulting behavior with the original example. Save failures as well as successes. If you repeat the evaluation, hold the prompt, tools and test inputs constant so that you can tell whether the model change helped.

Access, preview status and pricing

OpenRouter lists the model as stepfun/step-5-preview, with paid token-based access at $1.00 per million input tokens and $2.70 per million output tokens when checked. Context limits differ by provider (one lists about 262,000 tokens, the others about one million), so check the current endpoint’s context limit, supported inputs and parameters before sending a large document or video. The model card links deployment resources and weights; those are separate from buying hosted API usage.

Keep the first integration narrow. Record the selected provider and model identifier, set a spending limit and retain a known working configuration before changing a production workflow.

What to compare against

For a short text-only coding task, GPT-6 Luna lists lower per-token rates on OpenRouter: $0.10 input and $0.50 output per million tokens, against $1.00 and $2.70 for Step 5 Preview. OpenRouter lists Luna with text, image and file inputs, so the input difference to check is video, which it lists for Step 5 Preview but not for Luna. Compare completion quality and total task cost, not just a benchmark position.