Sparse MoE architecture
The sparse Mixture-of-Experts model has 600 billion total parameters and activates 27 billion parameters per token, reducing the amount of computation used for each token relative to a fully active model.
Step 5 Preview
Step 5 Preview is StepFun’s flagship AI model for agentic software engineering, professional knowledge work, and finance tasks, with vision input, a 1M-token context window, and API access.
Step 5 Preview overview
Step 5 Preview is a flagship AI model from StepFun for agentic work, software engineering, and professional knowledge tasks. It combines a sparse Mixture-of-Experts architecture with 600B total parameters, 27B active per token, a 1M-token context window, and vision input.
The model is available through StepFun products and API, and the source says open weights are planned for release on October 15. Its published evaluations cover coding, agentic tasks, professional knowledge work, and finance, while its demonstrations show workflows that depend on execution, feedback, and iterative revision rather than one-shot text generation.
Key features
Long-context processing
A 1M-token context window supports work across extensive codebases, long documents, and other large working contexts.
Vision-enabled development
Vision input and examples such as photo-to-3D room planning extend the model beyond text-only software tasks into visual application development.
Broad software engineering coverage
The model covers bug repair, feature modification, refactoring, documentation, performance tuning, code generation, CI/CD operations, and other repository-level work across 33 programming languages.
Long-horizon execution
Step 5 Preview is evaluated on tasks that require repeated execution, feedback, and revision, including GPU-kernel optimization and automated post-training data-loop improvement.
Professional and financial work
The source reports capabilities in professional knowledge work with particular strength in finance, alongside results from the FrontierFinance and DRACO evaluations.
Use cases
Software engineering and code maintenance
Use the model for repository-level work such as feature changes, bug repair, refactoring, documentation, performance tuning, and CI/CD-related tasks. The source reports coverage across 553 repositories and 33 programming languages in StepCodeBench.
Hardware development and debugging
Give the model development documentation and authorization to work with programmable hardware. The product page describes use of cameras, COM ports, screenshots, and simulated mouse input during development and debugging.
Visual applications and 3D experiences
Apply the model to visually oriented development, such as turning a room photo into an editable 3D environment, refining Blender assets, and incorporating them into interactive web applications or games with Three.js.
Research and performance optimization
Use extended execution loops for technical optimization or model-development workflows. Demonstrated examples include optimizing an MLA GPU kernel through repeated measurement and improving a post-training data loop based on downstream results.
Finance and professional analysis
Support finance-oriented professional knowledge work and analysis. The page includes finance-specific evaluations, but it does not document a complete production trading or advisory workflow for Step 5 Preview.
Pros and Cons
Pros
- Supports a 1M-token context window and vision input.
- Covers a wide range of software engineering tasks across 33 programming languages and 20 application domains.
- Designed for multi-step work involving execution, feedback, and revision.
- Published evaluations include software engineering, agentic tasks, professional knowledge work, and finance.
- Available through StepFun products and API, with open-weight release stated as planned.
Cons
- The source does not provide verified pricing, quotas, billing details, or specific API limits.
- Although the model performs strongly on many evaluations, the page notes a meaningful gap to the frontier on the most challenging long-running tasks.
- Concrete setup, integration, and operational requirements are not described in the available product evidence.
FAQ
How can Step 5 Preview be accessed?
Step 5 Preview is available through StepFun products and its API. The source also states that open weights are planned for release on October 15, but it does not specify the year in the available page text.
What input capabilities does Step 5 Preview support?
The model supports a 1M-token context window and vision input, making it suitable for large codebases, extended documents, and visual development tasks. The available evidence does not specify file-size limits, output formats, or an exact API setup process.
What types of work is Step 5 Preview intended for?
The model is designed for software engineering, agentic tasks, professional knowledge work, and finance. It can also work with programmable hardware when given development documentation and user authorization.
Are there limitations to Step 5 Preview's performance?
The page reports strong results across several evaluations, but also notes that a meaningful gap to the frontier remains on the most challenging long-running software tasks. Results may vary by task, prompting, tools, and execution environment.
Quick Facts
- Product type
- Agentic AI and foundation model
- Model architecture
- Sparse Mixture of Experts
- Total parameters
- 600B
- Active parameters per token
- 27B
- Context window
- 1M tokens
- Input modality
- Text and vision
Step 5 Preview Alternatives
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
GitBot
GitBot is a local browser workspace for reusable coding bots powered by Claude Code, Codex, or OpenCode, with persistent threads and configurable permissions.
Grok 4.7
Grok 4.7 is an xAI model for coding and knowledge work, available via Grok Build, Cursor, the Grok API and more, from $2 per million input tokens.
AakarDev AI
AakarDev AI helps teams manage AI provider access, project setup, logs, and analytics in one dashboard. BYOK support included.
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
MeetStream
MeetStream is a meeting bot API for Zoom, Google Meet, Microsoft Teams, and Webex. Record, stream, and analyze meetings with usage-based pricing and $5 free credit.