StepFun has introduced Step 5 Preview, a flagship model designed for agentic tasks, including software engineering, financial analysis, and professional knowledge work. The model is capable of navigating extensive codebases, managing multi-file modifications, executing code, and iterating based on results. Additionally, it processes documents, charts, screenshots, and video to generate structured reports.
The model features a 600B parameter architecture with 27B active parameters per token and supports a 1M context window. According to data from Artificial Analysis via OpenRouter, the model achieves a score of 43.7 on the Intelligence Index, 88.3% on AA-LCR, and 46.5% on Humanity’s Last Exam. In financial benchmarking, it recorded 66.4% on FrontierFinance.
Step 5 Preview is a data-center model with an FP8 checkpoint size of 650.32 GB. It supports vLLM and SGLang with OpenAI-compatible APIs. For deployments on Blackwell hardware, the company utilizes the MegaMoE backend for expert-parallel modes. Recommended inference settings include a temperature of 1.0, top_p of 0.95, and top_k of 0.
The API pricing for the model is set at $1.00 per 1 million input tokens and $2.70 per 1 million output tokens, with cache hits priced at $0.10 per 1 million tokens. The API supports tool calling, JSON Schema output, and three levels of reasoning effort: low, medium, and high.
