# MMLU → Vals GPQA Diamond

Calibration · https://benchgap.net/calibration/mmlu/current/to/vals-gpqa-diamond/current · data retrieved 2026-11-18

Vals GPQA Diamond is estimated from MMLU with a Hill curve fitted on 8 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.85404^6.00 + x^6.00), R² = 0.73, cross-validated error 8.2 pp. It is used for 3 estimates.

| Estimated model | MMLU | Vals GPQA Diamond | Source |
|---|---|---|---|
| [LongCat-Flash-Lite-Sparse](https://benchgap.net/model/longcat-flash-lite-sparse.md) | 85.3% | 59.8% | estimated ± 8.2 pp, medium confidence |
| [Mixtral 8x22B Instruct v0.1](https://benchgap.net/model/mixtral-8x22b-instruct-v0-1.md) | 77.8% | 43.6% | estimated ± 8.2 pp, low confidence |
| [Gemma 2 27B](https://benchgap.net/model/gemma-2-27b.md) | 76.2% | 40.2% | estimated ± 8.2 pp, low confidence |

## Fitted on

| Model | MMLU | Vals GPQA Diamond |
|---|---|---|
| [Claude 3.5 Haiku](https://benchgap.net/model/claude-3-5-haiku.md) | 80.3% | 37.9% |
| [GPT-4.1](https://benchgap.net/model/gpt-4-1.md) | 90.2% | 65.4% |
| [GPT-4.1 mini](https://benchgap.net/model/gpt-4-1-mini.md) | 87.5% | 67.9% |
| [GPT-4.1 nano](https://benchgap.net/model/gpt-4-1-nano.md) | 80.1% | 50.8% |
| [GPT-4o mini](https://benchgap.net/model/gpt-4o-mini.md) | 82.0% | 44.2% |
| [o1](https://benchgap.net/model/o1.md) | 91.8% | 73.2% |
| [o3-mini](https://benchgap.net/model/o3-mini.md) | 86.9% | 75.5% |
| [o4-mini](https://benchgap.net/model/o4-mini.md) | 90.0% | 74.5% |
