Calibration
AA-HLE → HealthBench (raw)
HealthBench (raw) is estimated from AA-HLE with a Michaelis–Menten curve fitted on 7 models measured on both: y = 2.0000·x / (1.22997 + x), R² = 0.57, cross-validated error 6.4 pp. It is used for 188 estimates.
| Estimated model | AA-HLE | HealthBench (raw) | Source |
|---|---|---|---|
| A.X K2 | 29.6% | 38.8% | estimated ± 6.4 pp, low confidence |
| Apodex 1.1 | 34.1% | 43.4% | estimated ± 6.4 pp, low confidence |
| Apodex 1.1 Mini | 34.1% | 43.4% | estimated ± 6.4 pp, low confidence |
| Celeris-1 | 6.8% | 10.5% | estimated ± 6.4 pp, low confidence |
| Claude 3 Haiku | 4.1% | 6.5% | estimated ± 6.4 pp, low confidence |
| Claude 3 Opus | 2.8% | 4.5% | estimated ± 6.4 pp, low confidence |
| Claude 4.1 Opus Thinking | 12.5% | 18.5% | estimated ± 6.4 pp, low confidence |
| Claude 4 Sonnet | 4.3% | 6.8% | estimated ± 6.4 pp, low confidence |
| Claude Fable 5 | 55.5% | 62.2% | estimated ± 6.4 pp, low confidence |
| Claude Fable 5.1 | 59.1% | 64.9% | estimated ± 6.4 pp, low confidence |
| Claude Haiku 5.5 | 44.4% | 53.0% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.5 | 13.2% | 19.4% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.5 Thinking | 30.1% | 39.3% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.6 | 19.1% | 26.9% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 39.9% | 49.0% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.7 | 33.3% | 42.6% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 42.3% | 51.2% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.8 | 48.7% | 56.7% | estimated ± 6.4 pp, low confidence |
| Claude Sonnet 4.6 | 13.3% | 19.5% | estimated ± 6.4 pp, low confidence |
| Claude Sonnet 5 | 41.3% | 50.3% | estimated ± 6.4 pp, low confidence |
| Command A+ | 12.0% | 17.8% | estimated ± 6.4 pp, low confidence |
| DeepSeek-R1 | 15.8% | 22.8% | estimated ± 6.4 pp, low confidence |
| DeepSeek R1 Distill Qwen 32B | 4.6% | 7.2% | estimated ± 6.4 pp, low confidence |
| DeepSeek V3 | 2.9% | 4.6% | estimated ± 6.4 pp, low confidence |
| DeepSeek V3 0324 | 4.7% | 7.4% | estimated ± 6.4 pp, low confidence |
| DeepSeek V3.1 | 6.7% | 10.3% | estimated ± 6.4 pp, low confidence |
| DeepSeek V3.1 (Reasoning) | 14.3% | 20.8% | estimated ± 6.4 pp, low confidence |
| DeepSeek V3.2 | 11.2% | 16.7% | estimated ± 6.4 pp, low confidence |
| DeepSeek V4.1 Flash | 39.2% | 48.3% | estimated ± 6.4 pp, low confidence |
| DeepSeek V4 Flash 0731 | 38.6% | 47.8% | estimated ± 6.4 pp, low confidence |
| DeepSeek V4 Pro 0813 | 41.0% | 50.0% | estimated ± 6.4 pp, low confidence |
| Exaone 4.0 1.2B | 5.7% | 8.9% | estimated ± 6.4 pp, low confidence |
| Exaone 4.0 32B | 5.0% | 7.8% | estimated ± 6.4 pp, low confidence |
| Gemini 1.0 Pro | 4.2% | 6.6% | estimated ± 6.4 pp, low confidence |
| Gemini 1.5 Pro | 4.6% | 7.2% | estimated ± 6.4 pp, low confidence |
| Gemini 2.5 Flash | 4.7% | 7.4% | estimated ± 6.4 pp, low confidence |
| Gemini 2.5 Pro | 22.5% | 30.9% | estimated ± 6.4 pp, low confidence |
| Gemini 3.1 Pro | 47.0% | 55.3% | estimated ± 6.4 pp, low confidence |
| Gemini 3.5 Flash | 42.7% | 51.5% | estimated ± 6.4 pp, low confidence |
| Gemini 3.5 Flash-Lite | 18.8% | 26.5% | estimated ± 6.4 pp, low confidence |
| Gemini 3.6 Flash | 40.8% | 49.8% | estimated ± 6.4 pp, low confidence |
| Gemini 3.7 Flash | 47.9% | 56.1% | estimated ± 6.4 pp, low confidence |
| Gemini 3.8 Flash | 47.8% | 56.0% | estimated ± 6.4 pp, low confidence |
| Gemini 3 Flash | 15.0% | 21.7% | estimated ± 6.4 pp, low confidence |
| Gemini 3 Pro | 39.7% | 48.8% | estimated ± 6.4 pp, low confidence |
| Gemini 4 Argon | 57.1% | 63.4% | estimated ± 6.4 pp, low confidence |
| Gemma 3 27B | 4.4% | 6.9% | estimated ± 6.4 pp, low confidence |
| Gemma 4 12B | 15.7% | 22.6% | estimated ± 6.4 pp, low confidence |
| Gemma 4 26B A4B | 19.3% | 27.1% | estimated ± 6.4 pp, low confidence |
| Gemma 4 31B | 23.6% | 32.2% | estimated ± 6.4 pp, low confidence |
| Gemma 4 E2B | 4.8% | 7.5% | estimated ± 6.4 pp, low confidence |
| Gemma 4 E4B | 3.8% | 6.0% | estimated ± 6.4 pp, low confidence |
| GLM-4.5-Air | 7.0% | 10.8% | estimated ± 6.4 pp, low confidence |
| GLM-4.6 | 5.5% | 8.6% | estimated ± 6.4 pp, low confidence |
| GLM-4.7 | 27.4% | 36.4% | estimated ± 6.4 pp, low confidence |
| GLM-5 | 29.3% | 38.5% | estimated ± 6.4 pp, low confidence |
| GLM-5.1 | 30.1% | 39.3% | estimated ± 6.4 pp, low confidence |
| GLM-5.2 | 41.1% | 50.1% | estimated ± 6.4 pp, low confidence |
| GLM-5.3 | 42.3% | 51.2% | estimated ± 6.4 pp, low confidence |
| GLM-5.3-Flash | 39.9% | 49.0% | estimated ± 6.4 pp, low confidence |
| GLM-5-Turbo | 27.8% | 36.9% | estimated ± 6.4 pp, low confidence |
| GLM-5V-Turbo | 17.1% | 24.4% | estimated ± 6.4 pp, low confidence |
| GPT-4.1 | 4.2% | 6.6% | estimated ± 6.4 pp, low confidence |
| GPT-4.1 mini | 5.0% | 7.8% | estimated ± 6.4 pp, low confidence |
| GPT-4.1 nano | 3.8% | 6.0% | estimated ± 6.4 pp, low confidence |
| GPT-4 Turbo | 3.1% | 4.9% | estimated ± 6.4 pp, low confidence |
| GPT-4o | 2.4% | 3.8% | estimated ± 6.4 pp, low confidence |
| GPT-4o mini | 4.2% | 6.6% | estimated ± 6.4 pp, low confidence |
| GPT-5.1 | 28.5% | 37.6% | estimated ± 6.4 pp, low confidence |
| GPT-5.1-Codex | 25.7% | 34.6% | estimated ± 6.4 pp, low confidence |
| GPT-5.1-Codex-Max | 25.7% | 34.6% | estimated ± 6.4 pp, low confidence |
| GPT-5.2 | 37.7% | 46.9% | estimated ± 6.4 pp, low confidence |
| GPT-5.2-Codex | 35.7% | 45.0% | estimated ± 6.4 pp, low confidence |
| GPT-5.3 Codex | 42.5% | 51.4% | estimated ± 6.4 pp, low confidence |
| GPT-5.4 | 43.7% | 52.4% | estimated ± 6.4 pp, low confidence |
| GPT-5.4 mini | 28.1% | 37.2% | estimated ± 6.4 pp, low confidence |
| GPT-5.4 nano | 28.3% | 37.4% | estimated ± 6.4 pp, low confidence |
| GPT-5.5 | 45.8% | 54.3% | estimated ± 6.4 pp, low confidence |
| GPT-5.6 Luna | 39.5% | 48.6% | estimated ± 6.4 pp, low confidence |
| GPT-5.6 Sol | 49.5% | 57.4% | estimated ± 6.4 pp, low confidence |
| GPT-5.6 Terra | 42.9% | 51.7% | estimated ± 6.4 pp, low confidence |
| GPT-5 (high) | 28.5% | 37.6% | estimated ± 6.4 pp, low confidence |
| GPT-5 (medium) | 25.4% | 34.2% | estimated ± 6.4 pp, low confidence |
| GPT-OSS 120B | 19.6% | 27.5% | estimated ± 6.4 pp, low confidence |
| GPT-OSS 20B | 11.0% | 16.4% | estimated ± 6.4 pp, low confidence |
| Granite-4.0-350M | 5.5% | 8.6% | estimated ± 6.4 pp, low confidence |
| Granite-4.0-H-1B | 5.0% | 7.8% | estimated ± 6.4 pp, low confidence |
| Granite-4.0-H-350M | 6.4% | 9.9% | estimated ± 6.4 pp, low confidence |
| Granite 4.2 30B | 11.2% | 16.7% | estimated ± 6.4 pp, low confidence |
| Granite 4.2 3B | 6.6% | 10.2% | estimated ± 6.4 pp, low confidence |
| Granite 4.2 8B | 9.7% | 14.6% | estimated ± 6.4 pp, low confidence |
| Grok 4 | 26.7% | 35.7% | estimated ± 6.4 pp, low confidence |
| Grok 4.1 Fast | 5.1% | 8.0% | estimated ± 6.4 pp, low confidence |
| Grok 4.1 Fast (Reasoning) | 19.3% | 27.1% | estimated ± 6.4 pp, low confidence |
| Grok 4.3 | 37.2% | 46.4% | estimated ± 6.4 pp, low confidence |
| Grok 4.5 | 42.7% | 51.5% | estimated ± 6.4 pp, low confidence |
| Grok 4.6 | 42.9% | 51.7% | estimated ± 6.4 pp, low confidence |
| Grok 4.7 | 43.1% | 51.9% | estimated ± 6.4 pp, low confidence |
| Grok 4 Fast (Reasoning) | 19.1% | 26.9% | estimated ± 6.4 pp, low confidence |
| Grok Code Fast 1 | 8.0% | 12.2% | estimated ± 6.4 pp, low confidence |
| Hy3 | 33.5% | 42.8% | estimated ± 6.4 pp, low confidence |
| Hy3 Preview | 33.5% | 42.8% | estimated ± 6.4 pp, low confidence |
| Inkling | 31.9% | 41.2% | estimated ± 6.4 pp, low confidence |
| Inkling-Small | 33.3% | 42.6% | estimated ± 6.4 pp, low confidence |
| K-Exaone | 13.9% | 20.3% | estimated ± 6.4 pp, low confidence |
| K-EXAONE 2.0 | 18.6% | 26.3% | estimated ± 6.4 pp, low confidence |
| Kimi K2.6 | 37.5% | 46.7% | estimated ± 6.4 pp, low confidence |
| Kimi K2 | 7.4% | 11.4% | estimated ± 6.4 pp, low confidence |
| Kimi K2.5 | 30.7% | 39.9% | estimated ± 6.4 pp, low confidence |
| Kimi K2.5 (Reasoning) | 30.7% | 39.9% | estimated ± 6.4 pp, low confidence |
| Kimi K2.7 Code | 35.0% | 44.3% | estimated ± 6.4 pp, low confidence |
| Kimi K3 | 46.9% | 55.2% | estimated ± 6.4 pp, low confidence |
| LFM2.5-2.6B | 6.2% | 9.6% | estimated ± 6.4 pp, low confidence |
| LFM2.5-8B-A1B | 6.9% | 10.6% | estimated ± 6.4 pp, low confidence |
| LFM2.5-VL-1.6B-Extract | 5.1% | 8.0% | estimated ± 6.4 pp, low confidence |
| Ling 2.6 Flash | 6.3% | 9.7% | estimated ± 6.4 pp, low confidence |
| Ling 3.0 Flash | 23.7% | 32.3% | estimated ± 6.4 pp, low confidence |
| Ling 3.0 Flash FP8 | 23.7% | 32.3% | estimated ± 6.4 pp, low confidence |
| Ling 3.0 Flash VL | 22.0% | 30.3% | estimated ± 6.4 pp, low confidence |
| Ling 3.0 Tiny | 9.3% | 14.1% | estimated ± 6.4 pp, low confidence |
| Ling 3.1 Flash | 39.4% | 48.5% | estimated ± 6.4 pp, low confidence |
| Llama 3.1 405B | 4.0% | 6.3% | estimated ± 6.4 pp, low confidence |
| Llama 4 Maverick | 4.9% | 7.7% | estimated ± 6.4 pp, low confidence |
| Llama 4 Scout | 3.8% | 6.0% | estimated ± 6.4 pp, low confidence |
| Mercury 2.5 | 11.8% | 17.5% | estimated ± 6.4 pp, low confidence |
| MiMo-V2.5-Pro | 35.7% | 45.0% | estimated ± 6.4 pp, low confidence |
| MiMo-V2.6-Flash | 35.1% | 44.4% | estimated ± 6.4 pp, low confidence |
| MiMo-V2.6-Pro | 49.4% | 57.3% | estimated ± 6.4 pp, low confidence |
| MiMo-V2-Flash | 8.6% | 13.1% | estimated ± 6.4 pp, low confidence |
| MiMo-V2-Omni | 22.1% | 30.5% | estimated ± 6.4 pp, low confidence |
| MiMo-V2-Pro | 30.4% | 39.6% | estimated ± 6.4 pp, low confidence |
| MiniCPM5-2B | 8.9% | 13.5% | estimated ± 6.4 pp, low confidence |
| MiniMax M2.7 | 29.6% | 38.8% | estimated ± 6.4 pp, low confidence |
| MiniMax M3 | 39.0% | 48.1% | estimated ± 6.4 pp, low confidence |
| Mistral Large 2 | 3.3% | 5.2% | estimated ± 6.4 pp, low confidence |
| Mistral Large 3 | 4.2% | 6.6% | estimated ± 6.4 pp, low confidence |
| Mistral Large 4 | 35.0% | 44.3% | estimated ± 6.4 pp, low confidence |
| Mistral Medium 3 | 4.1% | 6.5% | estimated ± 6.4 pp, low confidence |
| Mistral Medium 3.5 128B | 13.8% | 20.2% | estimated ± 6.4 pp, low confidence |
| Mistral Small 4 | 9.9% | 14.9% | estimated ± 6.4 pp, low confidence |
| Mistral Small 4 (Reasoning) | 9.9% | 14.9% | estimated ± 6.4 pp, low confidence |
| Muse Glimmer 30B | 22.0% | 30.3% | estimated ± 6.4 pp, low confidence |
| Muse Spark | 40.7% | 49.7% | estimated ± 6.4 pp, low confidence |
| Muse Spark 1.1 | 46.2% | 54.6% | estimated ± 6.4 pp, low confidence |
| Muse Spark 1.2 | 45.5% | 54.0% | estimated ± 6.4 pp, low confidence |
| Muse Spark 1.3 | 48.7% | 56.7% | estimated ± 6.4 pp, low confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 10.6% | 15.9% | estimated ± 6.4 pp, low confidence |
| Nemotron 3 Nano 30B | 11.4% | 17.0% | estimated ± 6.4 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 4.8% | 7.5% | estimated ± 6.4 pp, low confidence |
| Nemotron 3 Super 100B | 20.8% | 28.9% | estimated ± 6.4 pp, low confidence |
| Nemotron 3 Ultra | 28.4% | 37.5% | estimated ± 6.4 pp, low confidence |
| Nemotron Ultra 253B | 7.4% | 11.4% | estimated ± 6.4 pp, low confidence |
| North Mini Code | 11.1% | 16.6% | estimated ± 6.4 pp, low confidence |
| Nova Pro | 3.2% | 5.1% | estimated ± 6.4 pp, low confidence |
| o1 | 7.0% | 10.8% | estimated ± 6.4 pp, low confidence |
| o3 | 20.1% | 28.1% | estimated ± 6.4 pp, low confidence |
| o3-mini | 7.9% | 12.1% | estimated ± 6.4 pp, low confidence |
| Phi-4 | 3.8% | 6.0% | estimated ± 6.4 pp, low confidence |
| Phi-4 Multimodal Instruct | 5.0% | 7.8% | estimated ± 6.4 pp, low confidence |
| Quasar 438B | 18.7% | 26.4% | estimated ± 6.4 pp, low confidence |
| Qwen2.5 Coder 32B Instruct | 3.5% | 5.5% | estimated ± 6.4 pp, low confidence |
| Qwen3.5-122B-A10B | 25.2% | 34.0% | estimated ± 6.4 pp, low confidence |
| Qwen3.5-27B | 23.9% | 32.5% | estimated ± 6.4 pp, low confidence |
| Qwen3.5-35B-A3B | 21.0% | 29.2% | estimated ± 6.4 pp, low confidence |
| Qwen3.5 397B | 19.8% | 27.7% | estimated ± 6.4 pp, low confidence |
| Qwen3.5 397B (Reasoning) | 19.8% | 27.7% | estimated ± 6.4 pp, low confidence |
| Qwen3.6-27B | 23.1% | 31.6% | estimated ± 6.4 pp, low confidence |
| Qwen3.6-35B-A3B | 22.2% | 30.6% | estimated ± 6.4 pp, low confidence |
| Qwen 3.6 Max (preview) | 30.8% | 40.1% | estimated ± 6.4 pp, low confidence |
| Qwen3.6 Plus | 27.8% | 36.9% | estimated ± 6.4 pp, low confidence |
| Qwen3.7 Max | 40.5% | 49.5% | estimated ± 6.4 pp, low confidence |
| Qwen3.7 Plus | 35.6% | 44.9% | estimated ± 6.4 pp, low confidence |
| Qwen3.8-27B | 33.9% | 43.2% | estimated ± 6.4 pp, low confidence |
| Qwen3.8-Flash-Next | 38.0% | 47.2% | estimated ± 6.4 pp, low confidence |
| Qwen3.8 Max Preview | 43.1% | 51.9% | estimated ± 6.4 pp, low confidence |
| Qwen3 Max | 11.9% | 17.6% | estimated ± 6.4 pp, low confidence |
| Qwen3-Omni-30B-A3B-Instruct | 4.6% | 7.2% | estimated ± 6.4 pp, low confidence |
| Qwen3-Omni-30B-A3B-Thinking | 7.5% | 11.5% | estimated ± 6.4 pp, low confidence |
| Sarvam 105B | 11.0% | 16.4% | estimated ± 6.4 pp, low confidence |
| Sarvam 30B | 7.5% | 11.5% | estimated ± 6.4 pp, low confidence |
| Solar Pro 2 | 3.7% | 5.8% | estimated ± 6.4 pp, low confidence |
| Solar Pro 3 | 10.3% | 15.5% | estimated ± 6.4 pp, low confidence |
| Solar Pro 4 | 29.2% | 38.4% | estimated ± 6.4 pp, low confidence |
| Step 3.7 Flash | 21.4% | 29.6% | estimated ± 6.4 pp, low confidence |
| Step 5 Preview | 46.5% | 54.9% | estimated ± 6.4 pp, low confidence |
| Trinity-Large-Preview | 15.8% | 22.8% | estimated ± 6.4 pp, low confidence |
| Trinity-Large-Thinking | 15.8% | 22.8% | estimated ± 6.4 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 3.6% | 5.7% | estimated ± 6.4 pp, low confidence |