Step 5 Preview API Access

StepFun's new flagship for coding agents comes to Featherless on Oct 15, the day its open weights ship. Sign up now and run it from launch day.

Trusted by teams building on open models

Built for long-horizon work

Modality

Text + image input

Context Size

1M native

Concurrent Unit Cost

TBA

Model Size

600B (27B active)

Quantization

FP8

Tool Calling

Supported

Architecture

Sparse Mixture-of-Experts

What is Step 5 Preview?

Step 5 Preview is StepFun's new flagship, built for coding agents and long, multi-step work, with finance as a particular strength. It's a mixture-of-experts model with 600B parameters, 27B of them active per token. It reads text and images and has a native 1M-token context window. StepFun releases the open weights on October 15, and Featherless serves it from that day. Weights: stepfun-ai/Step-5-Preview on Hugging Face. For the architecture and training details, read StepFun's Inside Step 5 Preview.

How it compares

44 on the Artificial Analysis Intelligence Index. Claude Opus 5.5 scores 58 at its max reasoning setting.
About 1/6 the cost per task of Claude Opus 5.5, per Artificial Analysis.

What it's good at

Coding agents. Fixes bugs in codebases it hasn't seen before, builds and refactors features across files, and keeps debugging until the changes run and pass checks.
Frontend from a design. Turns a screenshot, mockup or PRD into a working web page, and can build a playable browser game from a folder of assets.
Finance work. Reads annual reports and operating data, breaks down revenue, profit, cash flow and risk, and hands back a report, an editable spreadsheet or a slide deck.

Run it on Featherless

The Featherless API is OpenAI-compatible and doesn't log prompts or completions. If you already run Step 5 Preview in Cline, Kilo Code, OpenCode or Hermes Agent, switching on October 15 takes three settings:

API key Your Featherless key
Base URL https://api.featherless.ai/v1
Model ID Posted on this page at launch

One Featherless key covers 40,000+ open models now and Step 5 Preview from October 15. Sign up for launch access

Dedicated & private cloud

Running Step 5 Preview in production or across a team?

Run it on a private, dedicated node on AMD's MI-series — one flat monthly cost, your own reserved capacity, Claude- and OpenAI-compatible. Tell us your workload and we'll size it.

Book a CallMost teams spending more than ~$7.5K/mo on coding AI come out cheaper on a dedicated node.

Cheaper because of the hardware

We run on AMD MI-series rather than Nvidia, and we've done the kernel to reach peak efficiency. A structural cost base, not a promotional rate.

Flat, forecastable cost

One fixed monthly price. Add agents and seats without watching a per-token meter.

Private & isolated

A node reserved for your org; your code, prompts, and context stay inside your trust boundary.

Dedicated throughput & SLA

Reserved capacity tuned to your workload, with support and uptime commitments.