# Vals MMLU-Pro → ARC-AGI-3

Calibration · https://benchgap.net/calibration/1558 · data retrieved 2026-10-29

ARC-AGI-3 is estimated from Vals MMLU-Pro with a inverse Michaelis–Menten curve fitted on 12 models measured on both: y = 0.00041·(x − 0.0000) / (0.0000 + 0.9173 − x), R² = 0.84, cross-validated error 9.0 pp. It is used for 5 estimates.

| Estimated model | Vals MMLU-Pro | ARC-AGI-3 | Source |
|---|---|---|---|
| [Gemini 3.1 Flash-Lite](https://benchgap.net/model/gemini-3-1-flash-lite.md) | 86.2% | 0.6% | estimated ± 9.0 pp, medium confidence |
| [GLM-5.3-Flash](https://benchgap.net/model/glm-5-3-flash.md) | 86.1% | 0.6% | estimated ± 9.0 pp, medium confidence |
| [Qwen3.5 Flash](https://benchgap.net/model/qwen3-5-flash.md) | 84.1% | 0.5% | estimated ± 9.0 pp, low confidence |
| [MiMo-V2.5](https://benchgap.net/model/mimo-v2-5.md) | 82.9% | 0.4% | estimated ± 9.0 pp, low confidence |
| [GLM-4.5](https://benchgap.net/model/glm-4-5.md) | 81.2% | 0.3% | estimated ± 9.0 pp, low confidence |

## Fitted on

| Model | Vals MMLU-Pro | ARC-AGI-3 |
|---|---|---|
| [Claude Opus 4.7](https://benchgap.net/model/claude-opus-4-7.md) | 89.9% | 0.2% |
| [Claude Opus 4.8](https://benchgap.net/model/claude-opus-4-8.md) | 89.6% | 1.5% |
| [Claude Opus 5](https://benchgap.net/model/claude-opus-5.md) | 91.6% | 30.2% |
| [Gemini 3.1 Pro](https://benchgap.net/model/gemini-3-1-pro.md) | 91.0% | 0.4% |
| [Gemini 3.8 Flash](https://benchgap.net/model/gemini-3-8-flash.md) | 90.2% | 10.4% |
| [GPT-5.5](https://benchgap.net/model/gpt-5-5.md) | 88.1% | 0.4% |
| [GPT-5.6 Luna](https://benchgap.net/model/gpt-5-6-luna.md) | 86.0% | 0.2% |
| [GPT-5.6 Sol](https://benchgap.net/model/gpt-5-6-sol.md) | 89.1% | 7.8% |
| [GPT-5.6 Terra](https://benchgap.net/model/gpt-5-6-terra.md) | 86.7% | 0.8% |
| [Grok 4.20](https://benchgap.net/model/grok-4-20-beta.md) | 86.3% | 0.1% |
| [Grok 4.5](https://benchgap.net/model/grok-4-5.md) | 89.2% | 0.3% |
| [Grok 4.6](https://benchgap.net/model/grok-4-6.md) | 89.4% | 2.1% |