Calibration
BabaIsBench → Mystery Game Puzzles
Mystery Game Puzzles is estimated from BabaIsBench with a offset logistic curve fitted on 26 models measured on both: y = 0.2594 + (1.2000 − 0.2594) / (1 + exp(−65.50·(x − 1.0113))), R² = 0.61, cross-validated error 12.4 pp. It is used for 1 estimate.
| Estimated model | BabaIsBench | Mystery Game Puzzles | Source |
|---|---|---|---|
| Qwen3.8-27B | 21.0% | 25.9% | estimated ± 12.4 pp, low confidence |
Fitted on
| Model | BabaIsBench | Mystery Game Puzzles |
|---|---|---|
| Claude Fable 5 | 100.0% | 52.0% |
| Claude Fable 5.1 | 100.0% | 58.0% |
| Claude Opus 4.8 | 96.0% | 36.0% |
| Claude Opus 5 | 100.0% | 59.0% |
| Claude Sonnet 5 | 67.0% | 35.0% |
| DeepSeek V4 Flash 0731 | 75.0% | 34.0% |
| Gemini 3.1 Pro | 92.0% | 34.0% |
| Gemini 3.5 Flash | 88.0% | 32.0% |
| Gemini 3.6 Flash | 88.0% | 30.0% |
| Gemini 3.7 Flash | 100.0% | 37.0% |
| Gemini 3.8 Flash | 100.0% | 47.0% |
| GLM-5.2 | 96.0% | 15.0% |
| GLM-5.3 | 96.0% | 33.0% |
| GLM-5.3-Flash | 67.0% | 8.0% |
| GPT-5.5 | 100.0% | 56.0% |
| GPT-5.6 Luna | 63.0% | 21.0% |
| GPT-5.6 Sol | 100.0% | 58.0% |
| GPT-5.6 Terra | 88.0% | 35.0% |
| GPT-6 Astra | 100.0% | 84.0% |
| Grok 4.6 | 96.0% | 34.0% |
| Kimi K3 | 96.0% | 26.0% |
| MiniMax M3 | 46.0% | 8.0% |
| Muse Spark 1.3 | 46.0% | 25.0% |
| Qwen3.6-27B | 29.0% | 7.0% |
| Qwen3.7 Max | 25.0% | 32.0% |
| Qwen3.8 Max | 58.0% | 38.0% |