# ARC-AGI-2 → LABBench2

Calibration · https://benchgap.net/calibration/arc-agi-2/current/to/labbench2/current · data retrieved 2026-11-21

LABBench2 is estimated from ARC-AGI-2 with a Michaelis–Menten curve fitted on 6 models measured on both: y = 1.4452·x / (0.63446 + x), R² = 0.30, cross-validated error 3.1 pp. It is used for 1 estimate.

| Estimated model | ARC-AGI-2 | LABBench2 | Source |
|---|---|---|---|
| [dots3-note Preview](https://benchgap.net/model/dots3-note-preview.md) | 81.4% | 81.2% | estimated ± 3.1 pp, low confidence |

## Fitted on

| Model | ARC-AGI-2 | LABBench2 |
|---|---|---|
| [Claude Opus 5](https://benchgap.net/model/claude-opus-5.md) | 90.4% | 84.2% |
| [Gemini 3.7 Flash](https://benchgap.net/model/gemini-3-7-flash.md) | 84.6% | 82.1% |
| [Gemini 3.8 Flash](https://benchgap.net/model/gemini-3-8-flash.md) | 89.2% | 86.2% |
| [Gemini 4 Argon](https://benchgap.net/model/gemini-4-argon.md) | 91.3% | 88.8% |
| [GPT-5.6 Sol](https://benchgap.net/model/gpt-5-6-sol.md) | 92.5% | 82.1% |
| [GPT-5.6 Terra](https://benchgap.net/model/gpt-5-6-terra.md) | 82.5% | 81.2% |
