StepFun
availableShows if the model has enough results for an index.Step 5 Preview
Step 5 Preview is a reasoning model from StepFun in the Step 5 family. 24 benchmarks count toward its score, in 5 categories.
IndexOverall score out of 100.72.5 ±5.3
CoverageShare of the index weight with results.80%
SpeedOutput tokens per second.83/s
Input / 1MUS dollars per 1M input tokens.$1
Output / 1MUS dollars per 1M output tokens.$2.7
ContextMaximum tokens in one request.1M
EloLMArena rating and rank.N/A
The index is a score out of 100. The ± range shows how much it can change.
CapabilitiesScore per category, out of 100.
Out of 100Results
24 counted| BenchmarkThe test name. | CategoryThe capability that the test measures. | ResultThe score from the publisher. | IndexThis result as a score out of 100. | RunThe settings of the run. | DateDate of the result. | Published byThe source of the result. |
|---|---|---|---|---|---|---|
| GPQA Diamond | Knowledge | 93.5% | 64.6 | — | — | David Rein et al. |
| BrowseComp | Agentic | 88.7% | 74.2 | — | — | OpenAI |
| Artificial Analysis Long Context Reasoning | Reasoning | 88.3% | 69.3 | — | — | Artificial Analysis |
| MCP Atlas | Agentic | 85.6% | 72.6 | — | — | OpenAI |
| Terminal-Bench 2.1 (provider run) | Agentic | 85.0% | 74.2 | — | — | DeepSeek-AI |
| Terminal-Bench 2.1 (provider run) | Agentic | 85.0% | 74.2 | — | — | DeepSeek-AI |
| CyberGym | Agentic | 84.7% | 74.1 | — | — | Zhun Wang et al. |
| Data Research and Analysis with Complex Operations | Agentic | 83.3% | — | — | — | Anthropic |
| ProgramBench: Can Language Models Rebuild Programs From Scratch? | Coding | 80.5% | 77.1 | — | — | John Yang et al. |
| Artificial Analysis MMMU-Pro | Multimodal | 76.4% | 60.5 | — | — | Artificial Analysis |
| Massive Multi-discipline Multimodal Understanding Pro | Multimodal | 76.0% | 51.0 | — | — | MMMU-Pro authors |
| Toolathlon-Verified | Agentic | 74.1% | 72.3 | — | — | Moonshot AI |
| SWE-Marathon | Coding | 72.7% | — | — | — | Abundant AI and BenchFlow |
| DeepSWE | Agentic | 67.7% | 73.0 | — | — | Datacurve AI |
| OfficeQA Pro | Multimodal | 60.3% | 72.6 | — | — | OfficeQA Pro authors |
| JobBench | Agentic | 59.0% | 74.7 | — | — | Yuetai Li et al. |
| Scientific Code Benchmark | Coding | 58.9% | 70.0 | — | — | Benchmark authors |
| Artificial Analysis SciCode | Coding | 58.9% | 74.6 | — | — | Artificial Analysis |
| GDPval-AA normalized | Agentic | 53.3% | 76.5 | — | — | Artificial Analysis |
| Humanity's Last Exam | Knowledge | 46.5% | 68.2 | — | — | Center for AI Safety et al. |
| Artificial Analysis Humanity's Last Exam | Knowledge | 46.5% | 75.0 | — | — | Artificial Analysis |
| AutomationBench | Agentic | 44.0% | 89.2 | — | — | Moonshot AI |
| Artificial Analysis Intelligence Index | Knowledge | 43.7% | 77.1 | — | — | Artificial Analysis |
| Artificial Analysis Omniscience Accuracy | Knowledge | 41.5% | 65.2 | — | — | Artificial Analysis |
| MLS-Bench Lite | Coding | 40.5% | — | — | — | MLS-Bench |
| APEX-Agents | Agentic | 37.8% | 70.8 | — | — | Moonshot AI / APEX-Agents benchmark authors |
| Agents' Last Exam | Agentic | 29.5% | 69.0 | — | — | DeepSeek-AI |
| Critical Physics Tasks | Reasoning | 20.9% | 82.2 | — | — | Artificial Analysis |
24 benchmarks count, from 25 of 28 results. A grey row does not count. Too few models took that benchmark.