StepFun Step 5 Preview: 600B Sparse Model With 1M-Token Context Lands on OpenRouter

StepFun's Step 5 Preview is now on OpenRouter, which lists it as the company's flagship model for agentic work.

The architecture is a sparse Mixture-of-Experts design with 27 billion active parameters out of 600 billion total. The context window is 1,000,000 tokens. Pricing is $1.00 per million prompt tokens and $2.70 per million completion tokens.

On quality, the listing says the model "performs strongly in software engineering and professional..." and then the text cuts off. The page gives no benchmark, no named test and no comparison model in what is available. "Strongly" is an adjective, not a score.

Context: the sparse design is the reason the headline number and the working number differ. Only 27 billion of the 600 billion parameters are active at a time, which is what the active/total split in the listing describes.

Two things are still open. The listing does not say when a final, non-preview Step 5 will arrive, and it does not include the performance details that the truncated sentence promises.

My take: a million tokens of context and a flagship label make for a tidy spec sheet, and the price is public. At least the price tag isn't playing hide-and-seek. The evidence for "agentic" is not. Show me the eval, not the vibe.