StepFun has quietly added a new flagship model to OpenRouter: Step 5 Preview. According to its OpenRouter listing, it is built for agentic work and uses a sparse Mixture-of-Experts (MoE) architecture with 27 billion active parameters out of 600 billion total. The model is priced at $1 per million input tokens and $2.70 per million output tokens.
The headline number is the context window: 1,000,000 tokens, with a maximum output of 64,000 tokens. That combination puts Step 5 Preview in the same conversation as other long-context and agent-oriented models, but at a price point that is competitive for high-volume API use. The listing is live now, and it is the first public sign of this specific StepFun model.
Step 5 Preview specs and pricing on OpenRouter
The OpenRouter page lists only a few hard numbers, but they are the ones developers care about most. Step 5 Preview is a sparse MoE model, meaning it activates a smaller slice of its total parameters per request, which usually helps control cost and latency.
At $1 and $2.70 per million tokens, Step 5 Preview sits in the mid-range of frontier API pricing. The 1M context window is the main selling point for teams building agents that need to keep long documents, codebases or conversation histories in memory.
- Model: Step 5 Preview by StepFun
- Architecture: sparse Mixture-of-Experts
- Parameters: 27B active / 600B total
- Context window: 1,000,000 tokens
