Calibration
Artificial Analysis Intelligence Index → C-Eval
C-Eval is estimated from Artificial Analysis Intelligence Index with a Michaelis–Menten curve fitted on 5 models measured on both: y = 1.0056·x / (0.02053 + x), R² = 0.73, cross-validated error 1.0 pp. It is used for 123 estimates.
| Estimated model | Artificial Analysis Intelligence Index | C-Eval | Source |
|---|---|---|---|
| Apodex 1.1 Mini | 26.4% | 93.3% | estimated ± 1.0 pp, medium confidence |
| Celeris-1 | 6.4% | 76.0% | estimated ± 1.0 pp, low confidence |
| Claude 3 Haiku | 5.6% | 73.5% | estimated ± 1.0 pp, low confidence |
| Claude 3 Opus | 8.7% | 81.4% | estimated ± 1.0 pp, low confidence |
| Claude 4.1 Opus | 18.6% | 90.5% | estimated ± 1.0 pp, medium confidence |
| Claude 4.1 Opus Thinking | 22.9% | 92.3% | estimated ± 1.0 pp, medium confidence |
| Claude 4 Sonnet | 16.6% | 89.5% | estimated ± 1.0 pp, low confidence |
| Claude Fable 5 | 49.6% | 96.6% | estimated ± 1.0 pp, low confidence |
| Claude Haiku 5.5 | 43.4% | 96.0% | estimated ± 1.0 pp, low confidence |
| Claude Opus 4.5 Thinking | 29.1% | 93.9% | estimated ± 1.0 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 32.0% | 94.5% | estimated ± 1.0 pp, low confidence |
| Claude Opus 4.7 | 30.9% | 94.3% | estimated ± 1.0 pp, low confidence |
| Claude Opus 5.5 | 57.6% | 97.1% | estimated ± 1.0 pp, low confidence |
| Claude Sonnet 5.5 | 56.0% | 97.0% | estimated ± 1.0 pp, low confidence |
| Command A+ | 22.5% | 92.2% | estimated ± 1.0 pp, medium confidence |
| DeepSeek-R1 | 13.1% | 87.0% | estimated ± 1.0 pp, low confidence |
| DeepSeek R1 Distill Qwen 32B | 8.4% | 80.8% | estimated ± 1.0 pp, low confidence |
| DeepSeek V3 0324 | 9.7% | 83.0% | estimated ± 1.0 pp, low confidence |
| DeepSeek V3.1 | 13.7% | 87.5% | estimated ± 1.0 pp, low confidence |
| DeepSeek V3.1 (Reasoning) | 13.5% | 87.3% | estimated ± 1.0 pp, low confidence |
| DeepSeek V3.2 | 16.0% | 89.1% | estimated ± 1.0 pp, low confidence |
| Exaone 4.0 1.2B | 5.2% | 72.2% | estimated ± 1.0 pp, low confidence |
| Exaone 4.0 32B | 6.3% | 75.9% | estimated ± 1.0 pp, low confidence |
| Gemini 1.0 Pro | 5.3% | 72.6% | estimated ± 1.0 pp, low confidence |
| Gemini 1.5 Pro | 7.9% | 79.8% | estimated ± 1.0 pp, low confidence |
| Gemini 2.5 Flash | 9.9% | 83.2% | estimated ± 1.0 pp, low confidence |
| Gemini 3.1 Pro | 29.7% | 94.1% | estimated ± 1.0 pp, low confidence |
| Gemini 3.5 Flash-Lite | 22.2% | 92.0% | estimated ± 1.0 pp, medium confidence |
| Gemini 3.6 Flash | 34.0% | 94.8% | estimated ± 1.0 pp, low confidence |
| Gemini 3.7 Flash | 39.1% | 95.5% | estimated ± 1.0 pp, low confidence |
| Gemini 3.8 Flash | 40.9% | 95.8% | estimated ± 1.0 pp, low confidence |
| Gemini 3 Flash | 17.9% | 90.2% | estimated ± 1.0 pp, low confidence |
| Gemini 3 Pro | 28.0% | 93.7% | estimated ± 1.0 pp, low confidence |
| Gemini 4 Argon | 52.6% | 96.8% | estimated ± 1.0 pp, low confidence |
| Gemma 3 27B | 4.9% | 70.7% | estimated ± 1.0 pp, low confidence |
| GLM-4.5-Air | 11.1% | 84.9% | estimated ± 1.0 pp, low confidence |
| GLM-4.6 | 14.9% | 88.4% | estimated ± 1.0 pp, low confidence |
| GLM-5.3 | 44.8% | 96.1% | estimated ± 1.0 pp, low confidence |
| GLM-5.3-Flash | 41.8% | 95.8% | estimated ± 1.0 pp, low confidence |
| GLM-5-Turbo | 26.6% | 93.4% | estimated ± 1.0 pp, medium confidence |
| GLM-5V-Turbo | 23.5% | 92.5% | estimated ± 1.0 pp, medium confidence |
| GPT-4 Turbo | 7.0% | 77.9% | estimated ± 1.0 pp, low confidence |
| GPT-4o | 8.4% | 80.9% | estimated ± 1.0 pp, low confidence |
| GPT-4o mini | 6.7% | 76.9% | estimated ± 1.0 pp, low confidence |
| GPT-5.1 | 24.7% | 92.9% | estimated ± 1.0 pp, medium confidence |
| GPT-5.1-Codex | 23.7% | 92.5% | estimated ± 1.0 pp, medium confidence |
| GPT-5.1-Codex-Max | 23.7% | 92.5% | estimated ± 1.0 pp, medium confidence |
| GPT-5.2-Codex | 28.5% | 93.8% | estimated ± 1.0 pp, low confidence |
| GPT-5.3 Codex | 32.5% | 94.6% | estimated ± 1.0 pp, low confidence |
| GPT-5 (high) | 23.0% | 92.3% | estimated ± 1.0 pp, medium confidence |
| GPT-5 (medium) | 22.9% | 92.3% | estimated ± 1.0 pp, medium confidence |
| GPT-6.1 Sol | 51.8% | 96.7% | estimated ± 1.0 pp, low confidence |
| GPT-6 Luna | 38.1% | 95.4% | estimated ± 1.0 pp, low confidence |
| GPT-6 Sol | 47.6% | 96.4% | estimated ± 1.0 pp, low confidence |
| GPT-OSS 120B | 11.6% | 85.4% | estimated ± 1.0 pp, low confidence |
| GPT-OSS 20B | 9.0% | 81.8% | estimated ± 1.0 pp, low confidence |
| Granite-4.0-350M | 4.8% | 70.6% | estimated ± 1.0 pp, low confidence |
| Granite-4.0-H-1B | 5.2% | 72.1% | estimated ± 1.0 pp, low confidence |
| Granite-4.0-H-350M | 4.8% | 70.6% | estimated ± 1.0 pp, low confidence |
| Grok 4 | 22.5% | 92.1% | estimated ± 1.0 pp, medium confidence |
| Grok 4.1 Fast | 11.3% | 85.1% | estimated ± 1.0 pp, low confidence |
| Grok 4.1 Fast (Reasoning) | 20.4% | 91.3% | estimated ± 1.0 pp, medium confidence |
| Grok 4.5 | 38.8% | 95.5% | estimated ± 1.0 pp, low confidence |
| Grok 4.6 | 44.3% | 96.1% | estimated ± 1.0 pp, low confidence |
| Grok 4.7 | 46.5% | 96.3% | estimated ± 1.0 pp, low confidence |
| Grok 4 Fast (Reasoning) | 17.9% | 90.2% | estimated ± 1.0 pp, low confidence |
| Grok Code Fast 1 | 14.1% | 87.7% | estimated ± 1.0 pp, low confidence |
| Hy3 | 25.3% | 93.0% | estimated ± 1.0 pp, medium confidence |
| K-Exaone | 14.4% | 88.0% | estimated ± 1.0 pp, low confidence |
| Kimi K2 | 12.7% | 86.6% | estimated ± 1.0 pp, low confidence |
| Kimi K2.7 Code | 25.8% | 93.1% | estimated ± 1.0 pp, medium confidence |
| LFM2.5-2.6B | 8.4% | 80.8% | estimated ± 1.0 pp, low confidence |
| LFM2.5-8B-A1B | 7.2% | 78.3% | estimated ± 1.0 pp, low confidence |
| LFM2.5-VL-1.6B-Extract | 4.8% | 70.6% | estimated ± 1.0 pp, low confidence |
| Ling 3.0 Flash VL | 24.6% | 92.8% | estimated ± 1.0 pp, medium confidence |
| Ling 3.0 Tiny | 11.1% | 84.8% | estimated ± 1.0 pp, low confidence |
| Ling 3.1 Flash | 41.1% | 95.8% | estimated ± 1.0 pp, low confidence |
| Llama 3.1 405B | 7.3% | 78.5% | estimated ± 1.0 pp, low confidence |
| Llama 4 Maverick | 10.0% | 83.4% | estimated ± 1.0 pp, low confidence |
| Llama 4 Scout | 8.1% | 80.2% | estimated ± 1.0 pp, low confidence |
| Mercury 2.5 | 12.3% | 86.2% | estimated ± 1.0 pp, low confidence |
| MiMo-V2.6-Flash | 37.9% | 95.4% | estimated ± 1.0 pp, low confidence |
| MiMo-V2.6-Pro | 46.3% | 96.3% | estimated ± 1.0 pp, low confidence |
| MiMo-V2-Omni | 23.9% | 92.6% | estimated ± 1.0 pp, medium confidence |
| MiMo-V2-Pro | 28.6% | 93.8% | estimated ± 1.0 pp, low confidence |
| MiniMax M2.7 | 22.8% | 92.2% | estimated ± 1.0 pp, medium confidence |
| MiniMax M3 | 29.2% | 94.0% | estimated ± 1.0 pp, low confidence |
| Mistral Large 2 | 7.6% | 79.1% | estimated ± 1.0 pp, low confidence |
| Mistral Large 3 | 9.3% | 82.3% | estimated ± 1.0 pp, low confidence |
| Mistral Large 4 | 38.4% | 95.5% | estimated ± 1.0 pp, low confidence |
| Mistral Medium 3 | 9.0% | 82.0% | estimated ± 1.0 pp, low confidence |
| Mistral Medium 3.5 128B | 14.2% | 87.8% | estimated ± 1.0 pp, low confidence |
| Mistral Small 4 | 11.3% | 85.1% | estimated ± 1.0 pp, low confidence |
| Mistral Small 4 (Reasoning) | 11.3% | 85.1% | estimated ± 1.0 pp, low confidence |
| Muse Glimmer 30B | 17.5% | 90.0% | estimated ± 1.0 pp, low confidence |
| Muse Spark 1.2 | 39.6% | 95.6% | estimated ± 1.0 pp, low confidence |
| Muse Spark 1.3 | 48.1% | 96.4% | estimated ± 1.0 pp, low confidence |
| Nemotron 3 Nano 30B | 8.9% | 81.7% | estimated ± 1.0 pp, low confidence |
| Nemotron 3 Super 100B | 12.8% | 86.7% | estimated ± 1.0 pp, low confidence |
| Nemotron Ultra 253B | 7.5% | 79.0% | estimated ± 1.0 pp, low confidence |
| North Mini Code | 9.9% | 83.3% | estimated ± 1.0 pp, low confidence |
| Nova Pro | 7.0% | 77.7% | estimated ± 1.0 pp, low confidence |
| o1-preview | 11.4% | 85.2% | estimated ± 1.0 pp, low confidence |
| o3 | 20.2% | 91.3% | estimated ± 1.0 pp, medium confidence |
| o3-pro | 21.9% | 91.9% | estimated ± 1.0 pp, medium confidence |
| Phi-4 | 5.9% | 74.7% | estimated ± 1.0 pp, low confidence |
| Phi-4 Multimodal Instruct | 5.8% | 74.3% | estimated ± 1.0 pp, low confidence |
| Quasar 438B | 26.7% | 93.4% | estimated ± 1.0 pp, medium confidence |
| Qwen2.5 Coder 32B Instruct | 6.7% | 77.1% | estimated ± 1.0 pp, low confidence |
| Qwen3.5 397B (Reasoning) | 21.4% | 91.8% | estimated ± 1.0 pp, medium confidence |
| Qwen3.8 Max Preview | 45.4% | 96.2% | estimated ± 1.0 pp, low confidence |
| Qwen3 Max | 15.6% | 88.9% | estimated ± 1.0 pp, low confidence |
| Qwen3-Omni-30B-A3B-Instruct | 6.0% | 75.0% | estimated ± 1.0 pp, low confidence |
| Qwen3-Omni-30B-A3B-Thinking | 7.8% | 79.5% | estimated ± 1.0 pp, low confidence |
| Sarvam 105B | 8.8% | 81.5% | estimated ± 1.0 pp, low confidence |
| Sarvam 30B | 6.6% | 76.6% | estimated ± 1.0 pp, low confidence |
| Solar Pro 2 | 7.0% | 77.8% | estimated ± 1.0 pp, low confidence |
| Solar Pro 3 | 7.8% | 79.6% | estimated ± 1.0 pp, low confidence |
| Solar Pro 4 | 28.2% | 93.7% | estimated ± 1.0 pp, low confidence |
| Step 3.7 Flash | 19.5% | 91.0% | estimated ± 1.0 pp, medium confidence |
| Trinity-Large-Preview | 10.8% | 84.5% | estimated ± 1.0 pp, low confidence |
| Trinity-Large-Thinking | 10.8% | 84.5% | estimated ± 1.0 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 7.7% | 79.3% | estimated ± 1.0 pp, low confidence |