Calibration

V* → Video-MME (with subtitle)

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

Video-MME (with subtitle) is estimated from V* with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (0.9005 − 0.0000)·x^6.00 / (0.48662^6.00 + x^6.00), R² = 0.92, cross-validated error 3.1 pp. It is used for 1 estimate.

Estimated model	V*	Video-MME (with subtitle)	Source
Step 3.7 Flash	95.3%	88.5%	estimated ± 3.1 pp, medium confidence