Vision Evals
Looking for rankings based on real user votes? See Arena Rankings
Optical character recognition benchmarks. Models read text from images across OCRBench (general OCR), a license-plate dataset, and a focused scene-text dataset.
What is OCR?
OCR evals measure how accurately models read text from images. We merge results from OCRBench (a 10% subset of the public benchmark), a license-plate recognition dataset, and a focused scene-text dataset into one combined score.
Methodology
Each model runs every OCR prompt and returns a text answer. The answer is graded pass/fail against the ground truth. Score = combined pass-rate across all three datasets.
Token usage & cost. Where shown, “output tokens” is the median per-prompt output count measured directly from each provider’s API response, and includes reasoning / thinking tokens, normalized across providers so the figure is comparable (for example, Gemini reports reasoning separately, and we add it into the output count). Input tokens include image tokens, which dominate and differ by model. “Est. cost / task” is that measured token usage multiplied by the model’s published per-1M pricing at the time of our last price sync, so it is an estimate on this benchmark, not a universal model cost. Figures come from a single evaluation run at low temperature; output for reasoning models can vary run to run. Models we haven’t measured (or that don’t expose token usage) show no token or cost figure rather than a zero.
Last evaluated: July 9, 2026
Frequently Asked Questions
OCRBench (a 10% subset, general OCR), a license-plate text recognition dataset, and Focused Scene OCR.
Each model is given the full set of OCR prompts across all three datasets. Each answer is pass/fail against the ground truth. Score is the combined percentage correct.