stepfun-ai/Step-5-Preview
Step 5 Preview is StepFun's flagship 600B total parameter (27B active) sparse Mixture-of-Experts (MoE) model, supporting up to 1 million tokens of context. It accepts text, images, and video, excelling in agentic work, software engineering, and professional knowledge tasks, with a particular strength in finance. This model is designed for sustained reasoning and tool use across complex, long-horizon workflows.
Loading preview...
Step 5 Preview: StepFun's Flagship Agentic Model
Step 5 Preview is StepFun's leading model, a sparse Mixture-of-Experts (MoE) architecture with 600 billion total parameters and 27 billion active parameters. It is engineered for advanced agentic workflows, supporting an extensive context window of up to 1 million tokens and handling multimodal inputs including text, images, and video. This model is designed for sustained reasoning and complex tool use across various domains.
Key Capabilities
- Software Engineering: Navigates large codebases, coordinates changes across files, runs and verifies code, and iterates on results for repository-level and visual development.
- Long-Horizon Agentic Execution: Capable of performing tasks requiring repeated tool calls, execution, verification, and revision, utilizing its long-context understanding throughout the process.
- Professional Knowledge Work: Combines research, multimodal understanding, and tool use to analyze extensive documents, charts, screenshots, images, and video, producing structured reports and analyses.
- Focus on Finance: Demonstrates particular strength in financial tasks, integrating quantitative reasoning with the analysis of financial data and supporting documents.
Good for
- Developing and maintaining complex software projects.
- Automating multi-step tasks that require continuous interaction and adaptation.
- In-depth analysis of diverse professional documents and data, including financial reports.
- Applications requiring advanced reasoning and tool-use capabilities over very long contexts.