Step 5 Preview
Step 5 Preview is StepFun's 600B sparse MoE reasoning model with 27B active parameters, 1M token context, image input, and extended chain-of-thought.
Model Overview
Capabilities, design details, and architectural traits
Step 5 Preview - StepFun's sparse MoE reasoning model
Step 5 Preview is a reasoning large language model from StepFun, previewed as the company's flagship. Its defining idea is a sparse mixture-of-experts (MoE) architecture that retains a very large total parameter pool.
The model works through problems with extended thinking, a chain-of-thought process where it reasons step by step before producing a final answer. This makes it verbose: it generates far more output tokens than the median reasoning model on the way to an answer, which is the direct tradeoff for its depth of reasoning.
Key differentiators
| Trait | Detail |
|---|---|
| Sparse MoE architecture | 600B total parameters with only 27B activated per inference |
| 1M token context window | Long enough to reason over full transcripts, codebases, and document stacks in one pass |
| Extended thinking | Chain-of-thought reasoning that works through problems step by step before answering |
| Multimodal input | Accepts text and image input, generates text output |
Where it fits
Step 5 Preview targets enterprise and infrastructure use cases, with strong results on agentic business operations, quantitative analysis, and root-cause analysis tasks. Its combination of a million-token context and image understanding positions it for teams that need deep reasoning over long documents and complex, multi-step work.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | StepFun | Anthropic | Anthropic |
| Release Date | September 18, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1 Best Input Pricing | $4 | $2 |
| Output Pricing | $2.70 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimage | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 43.7 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from StepFun
Other models by StepFun
Top AI Models
Leading alternatives by intelligence score
Claude Opus 5.5
Muse Spark 1.3
GPT-6 Astra
