StepFunopen-weight
Released: 2026-09-18Artificial AnalysisStep 5
OpenRouter ranked #7 foundation model processing 6.57T tokens weekly. Deployed by stepfun.
Intelligence Index
28.5/ 100
Artificial Analysis Intelligence Index (independent composite)Throughput Speed
128tok/sec
Output streaming throughputTime to First Answer
24s
Time to first answer token, including any thinkingCost to Run the Index
2.35USD
Artificial Analysis cost of the full Index runMeasured results · Artificial Analysis
Independent evaluations of Step 5. Values are reported as published; nothing here is estimated or derived. Snapshot 2026-11-15.
- Intelligence Index
- 28.5
- Cost to run the Index
- $2.35
- Time per task
- 14.8 min
- Omniscience index
- 16.4
- Reasoning model
- Yes
- Cached-input price
- $0.1 / 1M
Reasoning & knowledge
- AA-Omniscience Accuracy36.4%
- CritPt20.9%
- CritPt v2 (Hard)32.5%
- HLE-Diamond24.3%
- Humanity's Last Exam46.5%
Coding & terminal agents
- FrontierCode 1.138.4%
- FrontierCode-AA 2.012.0%
- SciCode58.9%
- Terminal-Bench 4.033.3%
- Terminal-Bench-Science 0.12.4%
Agentic work
- AA-AnalystAgent35.0%
- AA-Briefcase v1.1 (Elo)1424 Elo
- APEX-Agents-AA38.0%
- AutomationBench-AA51.0%
- EnterpriseOps-Gym-AA47.2%
- GDPval-AA (normalized Elo)1591 Elo
- ITBench-AA55.6%
- Multimodal Knowledge Agent (MKA)27.6%
Long context & documents
Multimodal & science
- AA-3D (Elo)859 Elo
- Code2Video-AA (Elo)821 Elo
- MMMU-Pro76.4%
Health & safety
Headline Evaluation Breakdown
Six independently measured evaluations (Artificial Analysis)Reasoning · CritPt v2 (Hard)32.5%
Coding & agents · Terminal-Bench 4.033.3%
Long context · AA-LCR88.3%
Knowledge · Humanity's Last Exam46.5%
Sourced Benchmark Evaluations (21)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| AA-Omniscience (Accuracy) | Active | 36.4% | 93 | 2026-11-15 | independent | Source ↗ |
| APEX-Agents-AA | Active | 38% | 93 | 2026-11-15 | independent | Source ↗ |
| ITBench-AA | Active | 55.6% | 93 | 2026-11-15 | independent | Source ↗ |
| EnterpriseOps-Gym-AA | Active | 47.2% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-AnalystAgent | Active | 35% | 93 | 2026-11-15 | independent | Source ↗ |
| AutomationBench-AA | Active | 51% | 93 | 2026-11-15 | independent | Source ↗ |
| Multimodal Knowledge Agent (MKA) | Active | 27.6% | 93 | 2026-11-15 | independent | Source ↗ |
| MLCR-AA (Medical Long Context Reasoning) | Active | 18.3% | 93 | 2026-11-15 | independent | Source ↗ |
| GDP.pdf | Active | 14.8% | 93 | 2026-11-15 | independent | Source ↗ |
| FrontierCode-AA 2.0 | Active | 12% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-Omniscience (Non-Hallucination) | Active | 68.7% | 93 | 2026-11-15 | independent | Source ↗ |
| FrontierCode | Active | 38.4% | 98 | 2026-11-15 | independent | Source ↗ |
| HLE-Diamond | Active | 24.3% | 93 | 2026-11-15 | independent | Source ↗ |
| CritPt v2 (Hard) | Active | 32.5% | 93 | 2026-11-15 | independent | Source ↗ |
| CritPt | Active | 20.9% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-LCR (Long Context Reasoning) | Active | 88.3% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench-Science 0.1 | Active | 2.4% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench 4.0 | Active | 33.3% | 93 | 2026-11-15 | independent | Source ↗ |
| SciCode | Active | 58.9% | 93 | 2026-11-15 | independent | Source ↗ |
| MMMU-Pro | Active | 76.4% | 93 | 2026-11-15 | independent | Source ↗ |
| Humanity's Last Exam | Active | 46.5% | 93 | 2026-11-15 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run step-5-preview
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "step-5-preview",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Step 5 represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.