Step-5 Preview

Step-5 Preview by StepFun

Step-5 Preview is a hosted preview foundation model from StepFun, made available through OpenRouter and notable for its 1,000,000-token context window. Its headline differentiator is scale: the model is designed for tasks where the input is not a short prompt but an entire corpus—large codebases, long research reports, legal records, multi-document knowledge packs, chat histories, or extensive operational logs. The provided discovery data identifies it as a newly added long-context model, released on 2026-10-08, with support for text generation, long-context processing, and reasoning-oriented use cases.

The model’s core value is its ability to keep far more information in scope than conventional chat models. A million-token context window can let users supply multiple books, large technical specifications, full repositories, or many related documents in a single request. This is especially useful for synthesis, comparison, retrieval-like question answering, summarization across many sources, and reasoning over dependencies between distant sections of text. In practice, long context does not automatically guarantee perfect recall, but it can reduce the need for manual chunking, external retrieval pipelines, or iterative prompting.

Key capabilities

Step-5 Preview is positioned as a text-generation and reasoning model. It should be suitable for drafting, summarization, analysis, structured extraction, document review, instruction following, and multi-step reasoning over very large inputs. Its long-context design makes it particularly relevant for workflows that require continuity across a large information space: tracing requirements through documentation, comparing many policy versions, reviewing transcripts, or asking questions over large technical archives.

Technical specifications

• Provider: StepFun
• Model name: Step-5 Preview
• Release date: 2026-10-08
• Availability: Hosted model available via OpenRouter
• Context window: 1,000,000 tokens
• Modalities: Text input and text output, based on the provided capability data
• Capabilities: Text generation, long-context processing, reasoning
• Max output: Not specified in the provided discovery data; users should verify the current OpenRouter listing
• Pricing: Not specified here and may vary by OpenRouter routing or provider policy
• Weights/license: No open-weight release is indicated; it should be treated as a hosted API model unless StepFun publishes otherwise
• Status: Preview, meaning behavior, pricing, limits, or availability may change

Strengths and benefits

The main benefit is breadth of context. For users who frequently hit context limits, Step-5 Preview can simplify prompting and preserve more source material in one request. It may also reduce engineering overhead for document splitting and retrieval orchestration. As a preview model, it may expose StepFun’s latest progress in long-context foundation model design before a more finalized release.

Limitations and caveats

The preview label is important. Output consistency, latency, rate limits, pricing, and maximum generation length may not be stable or fully documented. Very long prompts can also be expensive and slower to process, and million-token inputs may still suffer from attention dilution, missed details, or uneven recall. Users should validate answers against source material, especially for legal, financial, medical, or production decisions. Because no open weights are indicated, customization is likely limited to prompting and API-level integration rather than self-hosted fine-tuning.

Comparison

Compared with standard short- or medium-context models, Step-5 Preview’s differentiator is clearly its million-token context capacity. It competes more directly with other long-context hosted models that emphasize full-document or full-corpus reasoning. Compared with earlier or smaller StepFun offerings, the key advance appears to be expanded context length and preview access to a newer generation of foundation-model capabilities.