The Pulse
StepFun’s 600B Step 5 Preview Arrives on OpenRouter
StepFun’s Step 5 Preview is listed on OpenRouter with 600 billion total parameters, 27 billion active per token and a one-million-token context window. API access costs $1 per million input tokens, while StepFun says open weights are planne
AI.info Team · Written with AI ·
Step 5 Preview is our new flagship model for agentic work, delivering frontier-level performance across software engineering and professional knowledge work, with particular strength in finance.
StepFun, in its official announcement
StepFun’s Step 5 Preview is available through OpenRouter as a hosted model, with a one-million-token context window and a sparse mixture-of-experts design listing 600 billion total parameters and 27 billion active per token. OpenRouter identifies October 8, 2026, as the model’s release date. The listing describes it as aimed at extended agent tasks involving large codebases and documents.
StepFun’s 600B model is available by API
The 600-billion figure describes the model’s total parameters, not the amount used for every token. In a sparse mixture-of-experts system, StepFun says 27 billion parameters are active per token. That distinction matters for understanding the model’s scale: its total size is much larger than the portion activated for an individual step of generation.
StepFun’s announcement presents the model for software engineering, professional knowledge work and finance, and says it can handle long-horizon agent execution. The OpenRouter listing says it accepts text, images and video as input and returns text. It also supports tool calling and structured output through JSON Schema, features intended for applications that need models to act on information or return machine-readable results.
A million-token window, with a 64,000-token output limit
The one-million-token context window lets a request include a large body of material, such as lengthy documents or code. The same listing sets a separate ceiling of 64,000 completion tokens, so the context limit does not mean a model can generate a million-token answer. For teams evaluating it, the published limits describe the API’s capacity, not how consistently the model will retrieve details from every part of a very long prompt.
OpenRouter lists input at $1 per million tokens and output at $2.70 per million, with cached input reads priced at $0.05 per million. The separate input and output rates mean a workload’s bill depends not only on how much material it sends, but also on how much the model generates. The listing provides these rates alongside the context and output limits.
Hosted access is not the same as open weights
Despite the model’s appearance on an API marketplace, Step 5 Preview is not yet a downloadable open-weight model. StepFun’s announcement says the weights are planned for October 15, 2026; as of October 9, that date is still in the future. The available product is a hosted preview that users can call through StepFun’s services and OpenRouter.
That distinction sets the immediate terms of access: developers can test the model through a paid API, but cannot yet inspect or run the announced weights themselves. StepFun’s capability descriptions and performance claims remain the company’s own positioning; the OpenRouter listing supplies specifications, access options and prices, not an independent evaluation. For now, the concrete offer is a hosted model with a one-million-token context, 64,000-token output ceiling and listed API rates of $1 per million input tokens and $2.70 per million output tokens.