- Step 5 Preview is StepFun’s flagship MoE model featuring 27B active and 600B total parameters.
- The model boasts a massive 1 million token context window and is highly optimized for developer agentic workflows and finance tasks.
- Available on OpenRouter, the API is priced competitively at $1.00 per million input and $2.70 per million output tokens.
AI developers and open-source enthusiasts have a powerful new tool to experiment with. Step 5 Preview, the flagship model from AI startup StepFun, has officially debuted on OpenRouter. Built on a sparse Mixture-of-Experts (MoE) architecture boasting a massive 600 billion total parameters (with 27 billion active), this model is designed specifically to handle heavy-duty, multi-step agentic workflows that require deep reasoning and tool integration.
What makes Step 5 Preview stand out in an increasingly crowded AI ecosystem is its massive 1.0M-token context window, paired with a capacity for up to 64,000 completion tokens. According to integration data, the model excels in software engineering, professional knowledge tasks, and complex financial analysis. It is purpose-built for extended tasks that require navigating sprawling codebases, processing massive documents, and systematically refining output over multiple steps.
OpenRouter’s integration benchmarks highlight impressive performance stats, with Step 5 Preview delivering a throughput of 65 tokens per second and a swift 1.49-second latency. On the pricing front, the model is highly accessible: input tokens are priced at $1.00 per million, output tokens cost $2.70 per million, and cache reads sit at an ultra-low $0.05 per million. Popular open-source developer tools like Cline, OpenClaw, and Nous Research’s Hermes Agent are already driving massive token volumes to the model, signaling rapid adoption for autonomous coding and browser automation tasks.
Source: Original Coverage
Leave a Reply