Updated Sept 2026
Reasoning leaderboard
Composite R-score across six suites. Click any column to sort; use the filters to compare open and closed models. Lower hallucination rate is better.
| Rank | Model | Organization | Overall | Math | Code | Logic | Multi-step | Halluc. ↓ | 30d Δ |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | Anthropic | 89.2 | 91.4 | 90.1 | 88.7 | 86.9 | 4.1 | ▲ 0.8 |
| 2 | GPT-5.2 | OpenAI | 88.6 | 92.0 | 88.9 | 87.5 | 86.0 | 5.2 | ▲ 1.2 |
| 3 | Gemini 3 Pro | 87.9 | 89.8 | 87.2 | 89.0 | 85.6 | 4.8 | ▲ 0.5 | |
| 4 | Grok 4.1 | xAI | 85.1 | 86.3 | 84.0 | 86.8 | 83.3 | 5.9 | ▲ 0.4 |
| 5 | DeepSeek-R2Open | DeepSeek | 84.7 | 87.9 | 86.4 | 82.1 | 82.4 | 6.3 | ▲ 2.1 |
| 6 | GLM-5Open | Z.ai | 84.0 | 85.2 | 85.8 | 82.9 | 82.1 | 6.6 | ▲ 1.8 |
| 7 | Qwen3-MaxOpen | Alibaba | 83.4 | 84.7 | 83.1 | 83.0 | 82.8 | 7.0 | ▲ 0.9 |
| 8 | Kimi K2.5Open | Moonshot AI | 82.6 | 83.4 | 84.2 | 81.0 | 81.8 | 7.4 | ▲ 1.5 |
| 9 | Llama 4 MaverickOpen | Meta | 78.9 | 77.5 | 79.8 | 80.2 | 78.1 | 9.2 | ▲ 0.3 |
| 10 | Mistral Large 3 | Mistral AI | 77.2 | 76.1 | 78.4 | 78.9 | 76.6 | 9.8 | ▲ 0.6 |
10 / 10 models · Top 10 shown of 27 tracked. Scores are illustrative while the evaluation pipeline is finalized — ±95% bootstrap confidence intervals of ±0.9–1.6 omitted for readability.