Calibration
τ²-bench results → HLE w/ tools
HLE w/ tools is estimated from τ²-bench results with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 0.69404·(x − 0.1111) / (0.1111 + 2.0000 − x), R² = 0.52, cross-validated error 5.5 pp. It is used for 99 estimates.
| Estimated model | τ²-bench results | HLE w/ tools | Source |
|---|---|---|---|
| Claude Fable 5 | 98.5% | 53.9% | estimated ± 5.5 pp, low confidence |
| GLM-5-Turbo | 98.5% | 53.9% | estimated ± 5.5 pp, low confidence |
| GLM-5V-Turbo | 98.5% | 53.9% | estimated ± 5.5 pp, low confidence |
| GLM-4.7 | 95.9% | 51.1% | estimated ± 5.5 pp, low confidence |
| Qwen 3.6 Max (preview) | 95.9% | 51.1% | estimated ± 5.5 pp, low confidence |
| MiMo-V2-Pro | 95.0% | 50.1% | estimated ± 5.5 pp, low confidence |
| Claude Opus 4.8 | 94.4% | 49.5% | estimated ± 5.5 pp, low confidence |
| Mistral Medium 3.5 128B | 94.2% | 49.3% | estimated ± 5.5 pp, low confidence |
| Qwen3.6-27B | 94.2% | 49.3% | estimated ± 5.5 pp, low confidence |
| Qwen3.5-27B | 93.9% | 49.0% | estimated ± 5.5 pp, low confidence |
| Qwen3.5-122B-A10B | 93.6% | 48.7% | estimated ± 5.5 pp, low confidence |
| Grok 4.1 Fast (Reasoning) | 93.3% | 48.4% | estimated ± 5.5 pp, low confidence |
| GPT-5.2-Codex | 92.1% | 47.2% | estimated ± 5.5 pp, low confidence |
| MiMo-V2-Omni | 91.2% | 46.4% | estimated ± 5.5 pp, low confidence |
| Kimi K2.7 Code | 90.1% | 45.3% | estimated ± 5.5 pp, low confidence |
| Trinity-Large-Preview | 90.1% | 45.3% | estimated ± 5.5 pp, low confidence |
| Trinity-Large-Thinking | 90.1% | 45.3% | estimated ± 5.5 pp, low confidence |
| Claude Opus 4.5 Thinking | 89.5% | 44.7% | estimated ± 5.5 pp, low confidence |
| Qwen3.5-35B-A3B | 89.2% | 44.5% | estimated ± 5.5 pp, low confidence |
| MiniMax M3 | 88.9% | 44.2% | estimated ± 5.5 pp, low confidence |
| LFM2.5-8B-A1B | 88.1% | 43.4% | estimated ± 5.5 pp, low confidence |
| Gemini 3 Pro | 87.1% | 42.5% | estimated ± 5.5 pp, low confidence |
| GPT-5 (medium) | 86.5% | 42.0% | estimated ± 5.5 pp, low confidence |
| Solar Pro 3 | 86.3% | 41.8% | estimated ± 5.5 pp, low confidence |
| GPT-5.3 Codex | 86.0% | 41.5% | estimated ± 5.5 pp, low confidence |
| Ling 2.6 Flash | 86.0% | 41.5% | estimated ± 5.5 pp, low confidence |
| Command A+ | 85.0% | 40.7% | estimated ± 5.5 pp, low confidence |
| GPT-5.2 | 84.8% | 40.5% | estimated ± 5.5 pp, low confidence |
| GPT-5 (high) | 84.8% | 40.5% | estimated ± 5.5 pp, low confidence |
| MiMo-V2-Flash | 83.9% | 39.7% | estimated ± 5.5 pp, low confidence |
| Qwen3.5 397B (Reasoning) | 83.9% | 39.7% | estimated ± 5.5 pp, low confidence |
| GPT-5.1-Codex | 83.0% | 38.9% | estimated ± 5.5 pp, low confidence |
| GPT-5.1-Codex-Max | 83.0% | 38.9% | estimated ± 5.5 pp, low confidence |
| GPT-5.1 | 81.9% | 38.0% | estimated ± 5.5 pp, low confidence |
| o3 | 80.7% | 37.0% | estimated ± 5.5 pp, low confidence |
| LLaDA2.2-flash | 80.3% | 36.7% | estimated ± 5.5 pp, low confidence |
| Ternary Bonsai 2 27B | 80.2% | 36.6% | estimated ± 5.5 pp, low confidence |
| DeepSeek V3.2 | 78.9% | 35.6% | estimated ± 5.5 pp, low confidence |
| Agents-A1-4B | 78.2% | 35.0% | estimated ± 5.5 pp, low confidence |
| GLM-4.6 | 76.9% | 34.0% | estimated ± 5.5 pp, low confidence |
| Grok Code Fast 1 | 75.7% | 33.1% | estimated ± 5.5 pp, low confidence |
| Grok 4 | 74.9% | 32.5% | estimated ± 5.5 pp, low confidence |
| K-Exaone | 74.3% | 32.1% | estimated ± 5.5 pp, low confidence |
| Qwen3 Max | 74.3% | 32.1% | estimated ± 5.5 pp, low confidence |
| Claude Opus 4.7 | 74.0% | 31.8% | estimated ± 5.5 pp, low confidence |
| Claude 4.1 Opus Thinking | 71.4% | 30.0% | estimated ± 5.5 pp, low confidence |
| Grok 4 Fast (Reasoning) | 65.8% | 26.1% | estimated ± 5.5 pp, low confidence |
| Grok 4.1 Fast | 63.7% | 24.8% | estimated ± 5.5 pp, low confidence |
| o1 | 62.6% | 24.1% | estimated ± 5.5 pp, low confidence |
| Kimi K2 | 61.1% | 23.1% | estimated ± 5.5 pp, low confidence |
| Gemma 4 31B | 59.9% | 22.4% | estimated ± 5.5 pp, low confidence |
| LLaDA2.2-mini | 57.5% | 21.0% | estimated ± 5.5 pp, low confidence |
| Gemini 2.5 Pro | 54.1% | 19.0% | estimated ± 5.5 pp, low confidence |
| GPT-4.1 mini | 52.9% | 18.3% | estimated ± 5.5 pp, low confidence |
| Claude 4 Sonnet | 52.3% | 18.0% | estimated ± 5.5 pp, low confidence |
| DeepSeek V3 0324 | 47.1% | 15.2% | estimated ± 5.5 pp, low confidence |
| GPT-4.1 | 47.1% | 15.2% | estimated ± 5.5 pp, low confidence |
| Sarvam 105B | 46.8% | 15.1% | estimated ± 5.5 pp, low confidence |
| GLM-4.5-Air | 46.5% | 14.9% | estimated ± 5.5 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 45.3% | 14.3% | estimated ± 5.5 pp, low confidence |
| Gemma 4 26B A4B | 43.6% | 13.5% | estimated ± 5.5 pp, low confidence |
| Gemini 3 Flash | 43.3% | 13.3% | estimated ± 5.5 pp, low confidence |
| Mistral Small 4 | 41.2% | 12.3% | estimated ± 5.5 pp, low confidence |
| Mistral Small 4 (Reasoning) | 41.2% | 12.3% | estimated ± 5.5 pp, low confidence |
| Nemotron 3 Nano 30B | 40.9% | 12.1% | estimated ± 5.5 pp, low confidence |
| DeepSeek V3.1 (Reasoning) | 37.4% | 10.5% | estimated ± 5.5 pp, low confidence |
| North Mini Code | 37.4% | 10.5% | estimated ± 5.5 pp, low confidence |
| DeepSeek-R1 | 36.5% | 10.1% | estimated ± 5.5 pp, low confidence |
| Gemma 4 12B | 36.3% | 10.0% | estimated ± 5.5 pp, low confidence |
| DeepSeek V3.1 | 34.8% | 9.3% | estimated ± 5.5 pp, low confidence |
| Sarvam 30B | 34.5% | 9.2% | estimated ± 5.5 pp, low confidence |
| Solar Pro 2 | 31.9% | 8.1% | estimated ± 5.5 pp, low confidence |
| Mistral Large 2 | 30.7% | 7.5% | estimated ± 5.5 pp, low confidence |
| o3-mini | 28.7% | 6.7% | estimated ± 5.5 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 26.6% | 5.8% | estimated ± 5.5 pp, low confidence |
| GPT-4o | 25.1% | 5.2% | estimated ± 5.5 pp, low confidence |
| Mistral Large 3 | 24.6% | 5.0% | estimated ± 5.5 pp, low confidence |
| Mistral Medium 3 | 24.3% | 4.9% | estimated ± 5.5 pp, low confidence |
| DeepSeek V3 | 22.8% | 4.3% | estimated ± 5.5 pp, low confidence |
| Qwen3-Omni-30B-A3B-Thinking | 21.3% | 3.7% | estimated ± 5.5 pp, low confidence |
| Claude 3 Haiku | 21.1% | 3.7% | estimated ± 5.5 pp, low confidence |
| Gemma 4 E2B | 20.8% | 3.5% | estimated ± 5.5 pp, low confidence |
| Gemma 4 E4B | 20.8% | 3.5% | estimated ± 5.5 pp, low confidence |
| Exaone 4.0 1.2B | 20.5% | 3.4% | estimated ± 5.5 pp, low confidence |
| Granite-4.0-H-1B | 19.6% | 3.1% | estimated ± 5.5 pp, low confidence |
| Llama 3.1 405B | 19.0% | 2.9% | estimated ± 5.5 pp, low confidence |
| Llama 4 Maverick | 17.8% | 2.4% | estimated ± 5.5 pp, low confidence |
| GPT-4.1 nano | 17.3% | 2.2% | estimated ± 5.5 pp, low confidence |
| Qwen3-Omni-30B-A3B-Instruct | 16.4% | 1.9% | estimated ± 5.5 pp, low confidence |
| Llama 4 Scout | 15.5% | 1.6% | estimated ± 5.5 pp, low confidence |
| Gemini 2.5 Flash | 14.9% | 1.3% | estimated ± 5.5 pp, low confidence |
| Granite-4.0-H-350M | 14.6% | 1.2% | estimated ± 5.5 pp, low confidence |
| Nova Pro | 14.0% | 1.0% | estimated ± 5.5 pp, low confidence |
| Granite-4.0-350M | 13.2% | 0.7% | estimated ± 5.5 pp, low confidence |
| Nemotron Ultra 253B | 11.4% | 0.1% | estimated ± 5.5 pp, low confidence |
| Exaone 4.0 32B | 4.1% | 0.0% | estimated ± 5.5 pp, low confidence |
| Gemma 3 27B | 10.5% | 0.0% | estimated ± 5.5 pp, low confidence |
| LFM2.5-VL-1.6B-Extract | 8.5% | 0.0% | estimated ± 5.5 pp, low confidence |
| Phi-4 | 0.0% | 0.0% | estimated ± 5.5 pp, low confidence |