Calibration
BabyVision w/ Python → PerceptionBench
PerceptionBench is estimated from BabyVision w/ Python with a inverse Michaelis–Menten curve fitted on 5 models measured on both: y = 0.74680·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.84, cross-validated error 2.5 pp. It is used for 4 estimates.
| Estimated model | BabyVision w/ Python | PerceptionBench | Source |
|---|---|---|---|
| Glimpse | 95.4% | 68.1% | estimated ± 2.5 pp, medium confidence |
| DeepSeek V4.1 Flash 1017 | 92.5% | 64.3% | estimated ± 2.5 pp, medium confidence |
| MiniMax M3.1 Flash | 91.0% | 62.3% | estimated ± 2.5 pp, medium confidence |
| Qwen3.8-27B | 85.6% | 55.9% | estimated ± 2.5 pp, low confidence |
Fitted on
| Model | BabyVision w/ Python | PerceptionBench |
|---|---|---|
| Inkling 2 Mini | 88.8% | 57.5% |
| Inkling 2 Small | 93.7% | 64.7% |
| Kimi K3.1 Pro | 95.4% | 68.3% |
| Kimi K3 | 85.7% | 58.5% |
| Qwen3.8 Max | 91.3% | 63.5% |