StepFun's Step 5 Preview with 1M-token context arrives on Vercel AI Gateway

What happened

Vercel has added Step 5 Preview, the flagship model from Chinese AI developer StepFun, to the lineup of models reachable through its AI Gateway. According to the Vercel blog, the model is positioned for agentic coding, professional knowledge work and financial analysis — use cases that range from building and debugging applications to producing research and analytical reports.

The headline technical feature is a 1M-token context window. Step 5 Preview accepts both text and image input, which means a single request can combine an entire large codebase, a stack of documents, and screenshots or charts. That combination of long context and multimodal input is aimed at developers who want an agent to reason over everything relevant at once rather than chunking material across multiple calls.

How to use it

The model is addressed as stepfun/step-5-preview. Developers already integrated with the AI Gateway can simply swap the model identifier into existing calls.

For tooling that talks to models directly — Vercel names Claude Code, Codex and Cursor as examples — the setup path runs through the Vercel CLI. The steps outlined in the announcement are installing the latest CLI globally with npm i -g vercel@latest, running vercel ai-gateway setup, and then selecting stepfun/step-5-preview inside the coding agent. Vercel also points to its coding agents guide for fuller instructions, and a model playground is available for anyone who wants to try the model before wiring it into a project.

What the gateway adds

The availability says as much about Vercel's infrastructure play as it does about the model itself. According to the Vercel blog, AI Gateway offers a unified API for invoking models, tracking usage and cost, and configuring retries, failover and performance optimisations — with the stated goal of delivering higher uptime than any individual upstream provider.

The gateway also ships with built-in custom reporting, Zero Data Retention support, per-API-key budgets and routing rules. For teams weighing vendor risk, the routing and failover features mean a model choice like Step 5 Preview does not have to be a single point of failure: traffic can be steered or shifted if a provider degrades.

Commercially, Vercel says the gateway mirrors provider pricing with no markup and charges no platform fee on inference — including on Bring Your Own Key requests. That pricing stance matters for cost-sensitive teams, since gateways that sit between developers and model providers sometimes add a per-request premium.

Why it matters

Two threads make this announcement worth noting. First, it puts a 1M-token multimodal model within easy reach of a large developer audience. Vercel's platform is already embedded in many front-end and full-stack workflows, so adding a model geared toward agentic coding lowers the friction for developers who want long-context reasoning over big codebases without re-architecting their stack.

Second, it continues the shift from single-model integrations toward gateway-mediated access to many models. When one API key can reach StepFun's model alongside others, with unified billing, observability and failover, the marginal cost of evaluating a new model approaches zero. That dynamic pressures model providers to compete on quality and price rather than distribution lock-in, and it gives developers leverage to swap models as the landscape shifts.

For teams already running agents like Claude Code or Cursor, the practical takeaway is simple: with a CLI update and a model selection, Step 5 Preview becomes another option in the toolbox — one that can ingest a whole repository plus supporting documents and images in a single request.