Selective prediction on Classical Minds metacognitive battery (test)
95Baseline ScoreGemini 3 Flash
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Gemini 3 FlashTier=Valid2026.04 | 95 | 53.9 | 0.004 | 0.003 | 0.004 | |
| GLM-5Tier=Valid2026.04 | 94.3 | 58.7 | 0.024 | 0.022 | 0.023 | |
| Gemini 3.1 ProTier=Invalid2026.04 | 94.3 | 52.2 | 0.007 | 0.014 | 0 | |
| Gemini 2.5 FlashTier=Valid2026.04 | 93.9 | 57.9 | 0.016 | 0.026 | 0.034 | |
| Opus 4.6Tier=Valid2026.04 | 93.7 | 61.7 | 0.018 | 0.022 | 0.032 | |
| Sonnet 4.6Tier=Valid2026.04 | 92.7 | 71.7 | 0.039 | 0.04 | 0.05 | |
| Gemini 2.5 ProTier=Valid2026.04 | 92.6 | 56.1 | 0.024 | 0.02 | 0.017 | |
| Qwen 80B ThinkTier=Invalid2026.04 | 92.6 | 51.8 | 0.024 | 0.023 | 0.036 | |
| GPT-5.4Tier=Valid2026.04 | 92.4 | 64.6 | 0.033 | 0.038 | 0.046 | |
| Qwen Coder 480BTier=Valid2026.04 | 90.8 | 68.6 | 0.032 | 0.043 | 0.042 | |
| Qwen 235BTier=Valid2026.04 | 90.3 | 64.8 | 0.035 | 0.04 | 0.029 | |
| Qwen 80B InstTier=Valid2026.04 | 89.5 | 58.4 | 0.024 | 0.029 | 0.04 | |
| Claude Haiku 4.5Tier=Valid2026.04 | 89.1 | 65.7 | 0.04 | 0.041 | 0.044 | |
| DeepSeek V3.2Tier=Valid2026.04 | 88.5 | 65.1 | 0.015 | 0.03 | 0.057 | |
| GPT-5.4 miniTier=Valid2026.04 | 88.4 | 63.3 | 0.021 | 0.04 | 0.048 | |
| Gemma 3 27BTier=Valid2026.04 | 88 | 63.1 | 0.03 | 0.041 | 0.04 | |
| Gemma 3 12BTier=Indet.2026.04 | 85.7 | 61.5 | 0.029 | 0.045 | 0.044 | |
| DeepSeek-R1Tier=Invalid2026.04 | 85.3 | 3.1 | 0.036 | 0.06 | 0.143 | |
| GPT-5.4 nanoTier=Indet.2026.04 | 79.6 | 56.5 | 0.03 | 0.049 | 0.044 | |
| Gemma 3 1BTier=Indet.2026.04 | 60.3 | 48.3 | 0.003 | 0.009 | 0.126 |