Step 5 Preview
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Reasoning82.5
- 2Agentic & tool use78.5
- 3Coding77.6
- 4Multimodal#2474.6
- 5Knowledge70.6
Benchmark results (21)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Terminal-Bench 2.1 | 85.0 | #17 of 68 |
| BrowseComp | 88.7 | #9 of 50 |
| MCP Atlas | 85.6 | #4 of 42 |
| JobBench | 59.0 | #9 of 34 |
| CyberGym | 84.7 | #6 of 25 |
| AutomationBench | 44.0 | #12 of 25 |
| Toolathlon-Verified | 74.1 | #7 of 22 |
| Terminal-Bench 4.0 | 33.30 | #15 of 18 |
| Agents' Last Exam | 29.5 | #10 of 17 |
| Hermes Index | 32.8 | #14 of 16 |
| Hermes Bench | 69.2 | #12 of 16 |
| APEX-Agents | 37.8 | #3 of 8 |
| DRACO | 83.3 | #4 of 6 |
Coding
| DeepSWE | 67.7 | #23 of 50 |
| SciCode | 58.9 | #1 of 23 |
| ProgramBench | 80.5 | #4 of 15 |
| MLS-Bench Lite | 40.5 | #3 of 4 |
Multimodal
| MMMU-Pro | 76.0 | #29 of 48 |
| OfficeQA Pro | 60.3 | #8 of 13 |
Knowledge
| HLE | 46.5 | #22 of 69 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| List price (BenchLM) | $1.00 | $2.70 | $0.050 |
| OpenRouter stepfun/step-5-preview | $1.00 | $2.70 | $0.10 |
StepFun lists `step-5-preview` at $1.00 per million cache-miss input tokens, $0.05 per million cache-hit input tokens, and $2.70 per million output tokens. Output tokens include both the reasoning process and final answer. StepFun now publishes self-hostable FP8 and BF16 weights under StepFun Model License 1.0. These prices describe the hosted API; running the weights requires hardware.
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text | 1441 | #73 · 7.1K votes |
| WebDev (coding)(High) | 1567 | #39 · 4.7K votes |
| Vision | 1265 | #49 · 1.6K votes |
Open weights
From the Hugging Face repository.
- Parameters
- 604B
- Licence
- other
- Downloads (30 days)
- 1.8K
- Repository
- stepfun-ai/Step-5-Preview ↗
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Similar score
- Qwen3.8 Max69.7 · —
- GPT-5.569.2 · $11.25
- Gemini 3.8 Flash69.2 · $1.50
- Kimi K370.0 · $6.00
- Claude Opus 4.870.0 · $10.00