Calibration
τ²-bench results → BFCL v4
BFCL v4 is estimated from τ²-bench results with a Michaelis–Menten curve fitted on 7 models measured on both: y = 2.0000·x / (1.80631 + x), R² = 0.52, cross-validated error 8.5 pp. It is used for 63 estimates.
| Estimated model | τ²-bench results | BFCL v4 | Source |
|---|---|---|---|
| GLM-5-Turbo | 98.5% | 70.6% | estimated ± 8.5 pp, low confidence |
| GLM-5V-Turbo | 98.5% | 70.6% | estimated ± 8.5 pp, low confidence |
| Qwen 3.6 Max (preview) | 95.9% | 69.4% | estimated ± 8.5 pp, low confidence |
| Qwen3.5 397B (Reasoning) | 95.6% | 69.2% | estimated ± 8.5 pp, low confidence |
| MiMo-V2-Pro | 95.0% | 68.9% | estimated ± 8.5 pp, low confidence |
| Qwen3.5-27B | 93.9% | 68.4% | estimated ± 8.5 pp, low confidence |
| Grok 4.1 Fast (Reasoning) | 93.3% | 68.1% | estimated ± 8.5 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 92.1% | 67.5% | estimated ± 8.5 pp, low confidence |
| GPT-5.2-Codex | 92.1% | 67.5% | estimated ± 8.5 pp, low confidence |
| Qwen3.5-4B | 92.1% | 67.5% | estimated ± 8.5 pp, low confidence |
| MiMo-V2-Omni | 91.2% | 67.1% | estimated ± 8.5 pp, low confidence |
| Claude Opus 4.5 Thinking | 89.5% | 66.3% | estimated ± 8.5 pp, low confidence |
| Qwen3.5-35B-A3B | 89.2% | 66.1% | estimated ± 8.5 pp, low confidence |
| Gemini 3 Pro | 87.1% | 65.1% | estimated ± 8.5 pp, low confidence |
| GPT-5 (medium) | 86.5% | 64.8% | estimated ± 8.5 pp, low confidence |
| GPT-5.3 Codex | 86.0% | 64.5% | estimated ± 8.5 pp, low confidence |
| Claude Opus 4.6 | 84.8% | 63.9% | estimated ± 8.5 pp, low confidence |
| GPT-5.2 | 84.8% | 63.9% | estimated ± 8.5 pp, low confidence |
| GPT-5.1-Codex | 83.0% | 63.0% | estimated ± 8.5 pp, low confidence |
| GPT-5.1-Codex-Max | 83.0% | 63.0% | estimated ± 8.5 pp, low confidence |
| o3 | 80.7% | 61.8% | estimated ± 8.5 pp, low confidence |
| Ternary Bonsai 2 27B | 80.2% | 61.5% | estimated ± 8.5 pp, low confidence |
| Claude Sonnet 4.6 | 79.5% | 61.1% | estimated ± 8.5 pp, low confidence |
| DeepSeek V3.2 | 78.9% | 60.8% | estimated ± 8.5 pp, low confidence |
| Agents-A1-4B | 78.2% | 60.4% | estimated ± 8.5 pp, low confidence |
| GLM-4.6 | 76.9% | 59.7% | estimated ± 8.5 pp, low confidence |
| Grok Code Fast 1 | 75.7% | 59.1% | estimated ± 8.5 pp, low confidence |
| Grok 4 | 74.9% | 58.6% | estimated ± 8.5 pp, low confidence |
| Qwen3 Max | 74.3% | 58.3% | estimated ± 8.5 pp, low confidence |
| Claude Opus 4.7 | 74.0% | 58.1% | estimated ± 8.5 pp, low confidence |
| Claude 4 Opus Thinking | 73.4% | 57.8% | estimated ± 8.5 pp, low confidence |
| Claude 4.1 Opus Thinking | 71.4% | 56.7% | estimated ± 8.5 pp, low confidence |
| Grok 4 Fast (Reasoning) | 65.8% | 53.4% | estimated ± 8.5 pp, low confidence |
| Grok 4.1 Fast | 63.7% | 52.1% | estimated ± 8.5 pp, low confidence |
| o1 | 62.6% | 51.5% | estimated ± 8.5 pp, low confidence |
| Kimi K2 | 61.1% | 50.6% | estimated ± 8.5 pp, low confidence |
| o4-mini | 55.6% | 47.1% | estimated ± 8.5 pp, low confidence |
| Nemotron 3.5 Bolt E4B BF16 | 55.1% | 46.7% | estimated ± 8.5 pp, low confidence |
| Claude 3.7 Sonnet Thinking | 54.7% | 46.5% | estimated ± 8.5 pp, low confidence |
| Nemotron 3.5 Bolt E4B | 52.3% | 44.9% | estimated ± 8.5 pp, low confidence |
| Claude 4 Sonnet | 52.3% | 44.9% | estimated ± 8.5 pp, low confidence |
| Claude 3.7 Sonnet | 50.0% | 43.4% | estimated ± 8.5 pp, low confidence |
| GPT-4.1 | 47.1% | 41.4% | estimated ± 8.5 pp, low confidence |
| Sarvam 105B | 46.8% | 41.2% | estimated ± 8.5 pp, low confidence |
| GLM-4.5-Air | 46.5% | 40.9% | estimated ± 8.5 pp, low confidence |
| Gemini 3 Flash | 43.3% | 38.7% | estimated ± 8.5 pp, low confidence |
| DeepSeek V3.1 (Reasoning) | 37.4% | 34.3% | estimated ± 8.5 pp, low confidence |
| DeepSeek-R1 | 36.5% | 33.6% | estimated ± 8.5 pp, low confidence |
| DeepSeek V3.1 | 34.8% | 32.3% | estimated ± 8.5 pp, low confidence |
| Sarvam 30B | 34.5% | 32.1% | estimated ± 8.5 pp, low confidence |
| Solar Pro 2 | 31.9% | 30.0% | estimated ± 8.5 pp, low confidence |
| Mistral Large 2 | 30.7% | 29.1% | estimated ± 8.5 pp, low confidence |
| Gemini 2.0 Flash | 29.5% | 28.1% | estimated ± 8.5 pp, low confidence |
| o3-mini | 28.7% | 27.4% | estimated ± 8.5 pp, low confidence |
| Devstral Small | 28.4% | 27.2% | estimated ± 8.5 pp, low confidence |
| Magistral Small | 26.6% | 25.7% | estimated ± 8.5 pp, low confidence |
| Mistral Medium 3 | 24.3% | 23.7% | estimated ± 8.5 pp, low confidence |
| Magistral Medium | 23.1% | 22.7% | estimated ± 8.5 pp, low confidence |
| Llama 3.1 70B | 21.6% | 21.4% | estimated ± 8.5 pp, low confidence |
| Qwen3-Omni-30B-A3B-Thinking | 21.3% | 21.1% | estimated ± 8.5 pp, low confidence |
| Claude 3 Haiku | 21.1% | 20.9% | estimated ± 8.5 pp, low confidence |
| Exaone 4.0 1.2B | 20.5% | 20.4% | estimated ± 8.5 pp, low confidence |
| Granite-4.0-H-1B | 19.6% | 19.6% | estimated ± 8.5 pp, low confidence |
Fitted on
| Model | τ²-bench results | BFCL v4 |
|---|---|---|
| Kimi K2.5 | 95.9% | 68.3% |
| LFM2.5-8B-A1B | 88.1% | 49.7% |
| LLaDA2.2-flash | 80.3% | 60.8% |
| LLaDA2.2-mini | 57.5% | 47.7% |
| Qwen3.7 Max | 94.7% | 75.0% |
| Qwen3.7 Plus | 93.0% | 72.9% |
| Spark-X2.6-4B | 86.9% | 71.8% |