Control Calibration on Control
94.6AccuracyClaude Haiku 4.5
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Claude Haiku 4.5Tier=frontier, Mode=aware2026.05 | 94.6 | 95 | 94.9 | 88.6 | 87.4 | 99 | |
| Claude Sonnet 4.6Tier=frontier, Mode=aware2026.05 | 94.4 | 94.7 | 95 | 90.3 | 91.2 | 98 | |
| Claude Sonnet 4.6Tier=frontier, Mode=blind2026.05 | 91.9 | 92.2 | 80.4 | 90.8 | 91.1 | 95.5 | |
| gpt-5.4-miniTier=frontier, Mode=aware2026.05 | 90.9 | 91.2 | 88.3 | 75.7 | 84.1 | 98.6 | |
| gpt-5.4-miniTier=frontier, Mode=blind2026.05 | 89.9 | 90 | 82.5 | 87.2 | 89.6 | 94.6 | |
| Gemini 3.1 Flash LiteTier=frontier, Mode=aware2026.05 | 89.2 | 89.6 | 90.6 | 80.4 | 84.4 | 93.2 | |
| Qwen3.6-35B-A3BTier=open, Mode=aware2026.05 | 89 | 89.3 | 90.3 | 76.6 | 86.2 | 94.3 | |
| Claude Haiku 4.5Tier=frontier, Mode=blind2026.05 | 87.7 | 88.1 | 79.8 | 79.1 | 80.1 | 96.5 | |
| Gemma-4-26B-A4B-itTier=open, Mode=aware2026.05 | 87 | 87.6 | 88.3 | 75.7 | 82.3 | 95 | |
| Ministral-3-14BTier=open, Mode=aware2026.05 | 85.8 | 85.7 | 86.6 | 56.5 | 74.3 | 98.1 | |
| Gemini 3.1 Flash LiteTier=frontier, Mode=blind2026.05 | 84.2 | 84.1 | 81.9 | 84.1 | 90 | 87.8 | |
| Ministral-3-14BTier=open, Mode=blind2026.05 | 82.3 | 82.7 | 77.5 | 68.9 | 79.3 | 93.3 | |
| Gemma-4-26B-A4B-itTier=open, Mode=blind2026.05 | 79 | 79.2 | 80.9 | 78.9 | 82.9 | 94.4 | |
| Qwen3.6-35B-A3BTier=open, Mode=blind2026.05 | 78.3 | 77.1 | 82.8 | 72.5 | 89.2 | 92 | |
| gpt-oss-20BTier=open, Mode=aware2026.05 | 74.3 | 71.9 | 82.1 | 83.9 | 74.1 | 82.9 | |
| gpt-oss-20BTier=open, Mode=blind2026.05 | 71.7 | 66.3 | 82.5 | 87.6 | 74.8 | 86.1 |