Step 5 Preview, a 1M-context MoE from StepFun, shows up on OpenRouter

Models with 1 million token context windows are now accessible, enabling complex analysis of entire codebases and lengthy documents.

StepFun's new model, Step 5 Preview, is now available on OpenRouter, offering a 1-million token context window and a Mixture-of-Experts architecture with 27 billion active parameters out of 600 billion total. This model is designed for complex agentic tasks, particularly in software engineering, finance, and professional knowledge work, capable of processing extensive codebases and documents over multiple steps. It supports multimodal inputs, including text, images, and video, and outputs text. The pricing is set at $1 per million input tokens and $2.70 per million output tokens, with a cache read cost of $0.05 per million tokens. Step 5 Preview was released on October 8, 2026, and is hosted on OpenRouter, which directly forwards requests to the provider. OpenRouter also provides performance metrics such as throughput, latency, and uptime for the model.

The availability of Step 5 Preview with a 1 million token context window signifies a significant leap in the scale of tasks Large Language Models can handle. This capacity is crucial for agentic workflows that require understanding vast amounts of information, such as comprehensive code reviews or in-depth financial analysis over extended periods. The Mixture-of-Experts architecture, with a substantial total parameter count yet efficient active parameter usage, suggests a trend towards more specialized and performant models that can tackle demanding, multi-step professional workloads.

This development directly impacts developers and businesses engaged in complex software engineering or data analysis, offering tools to process and synthesize information previously unmanageable within a single LLM call. The reduced cost-per-token for such large contexts, coupled with its availability on a platform like OpenRouter, lowers the barrier to entry for deploying advanced AI agents. The market can anticipate further innovation in models designed for deep professional domain expertise and extended task execution, potentially accelerating autonomous agent capabilities.