olmOCR-BenchParseBenchfr-bench-pdf2mdGlobal average

Global average

Equal-weight mean of the three overall scores · Complete evaluations only

Global average score versus model sizeLower parameter count and higher overall score are better. Dashed line: Pareto frontier.556065707580850.712481632Model parameters (B) · Smaller is betterOverall score · Higher is betterMistral OCR 4 — 69.16 points — Size undisclosed — Reported resultsMistral OCR 4 · 69.2LightOnOCR-3-4B — 78.5 points — 4.0B — Output with post-processingLightOnOCR-3 4BLightOnOCR-3-0.8B — 76.9 points — 0.8B — Output with post-processingLightOnOCR-3 0.8BLightOnOCR-3-1B — 75.2 points — 1.0B — Output with post-processingLightOnOCR-3-1BInfinity-Parser2-Pro — 75.0 points — 35.1B — Reported resultsInfinity-Parser2-Prochandra-ocr-2 — 75.0 points — 4.0B — Output with post-processingchandra-ocr-2surya-ocr-2 — 67.4 points — 0.7B — Output with post-processingsurya-ocr-2dots.mocr — 66.3 points — 3.0B — Reported resultsdots.mocrjina-ocr-v1 — 60.9 points — 3.4B — Reported resultsjina-ocr-v1
External modelsLightOnOCR-2LightOnOCR-3Pareto frontier
Hover or focus a point for its exact checkpoint and scores.
Method and source

Smaller size and higher score are better. The frontier includes points that no other comparable point matches or improves on in both dimensions, with a strict improvement in at least one. Comparisons use source precision; displayed scores are rounded to one decimal. Parameter counts are those in the tables, including nominal backbone sizes where specified.

Average = (olmOCR overall including headers/footers + ParseBench five-category overall + fr-bench-pdf2md overall) / 3. Only models with comparable scores and sizes on all three benchmarks are included. LightOnOCR-3 averages use the same candidate-1 checkpoints across all three benchmarks. Only olmOCR outputs may be postprocessed; ParseBench and fr-bench-pdf2md outputs are evaluated as submitted. This is a derived summary, not a published leaderboard metric. Mistral OCR 4 is a horizontal score reference because its size is undisclosed; it is excluded from the size frontier. Its olmOCR score is the user-provided reported 85.20; ParseBench and fr-bench-pdf2md retain their existing selected scores.

Leaderboard source revision: 3576c5240602588f596f8790a3a7161a524e59ad. Values are pinned to this snapshot.

Show scores and output details
Plotted scores
ModelSize (B)OverallOutput / inputsFrontier
LightOnOCR-3-4B4.078.5olmOCR-Bench: 86.3 (Output with post-processing); ParseBench: 75.1 (Evaluated output); fr-bench-pdf2md: 74.1 (Evaluated output)✓
LightOnOCR-3-0.8B0.876.9olmOCR-Bench: 85.5 (Output with post-processing); ParseBench: 74.6 (Evaluated output); fr-bench-pdf2md: 70.5 (Evaluated output)✓
LightOnOCR-3-1B1.075.2olmOCR-Bench: 84.5 (Output with post-processing); ParseBench: 71.4 (Evaluated output); fr-bench-pdf2md: 69.6 (Evaluated output)—
Infinity-Parser2-Pro35.175.0olmOCR-Bench: 87.6 (Reported results); ParseBench: 74.3 (Reported results); fr-bench-pdf2md: 63.2 (Reported results)—
chandra-ocr-24.075.0olmOCR-Bench: 85.8 (Output with post-processing); ParseBench: 70.1 (Evaluated output); fr-bench-pdf2md: 69.0 (Evaluated output)—
Mistral OCR 4Undisclosed69.2olmOCR-Bench: 85.2 (Reported results); ParseBench: 68.2 (Reported results); fr-bench-pdf2md: 54.1 (Reported results)Size undisclosed
surya-ocr-20.767.4olmOCR-Bench: 83.3 (Output with post-processing); ParseBench: 64.8 (Evaluated output); fr-bench-pdf2md: 54.1 (Evaluated output)✓
dots.mocr3.066.3olmOCR-Bench: 83.9 (Evaluated output); ParseBench: 55.8 (Evaluated output); fr-bench-pdf2md: 59.3 (Evaluated output)—
jina-ocr-v13.460.9olmOCR-Bench: 83.4 (Evaluated output); ParseBench: 45.9 (Evaluated output); fr-bench-pdf2md: 53.2 (Evaluated output)—