Official benchmarks map / ParseBench
KDLAI/KDL-Frontier-Parser-nano-v2 is ranked first on the ParseBench leaderboard with 78.96, 5.0% ahead of nvidia/Nemotron-3.5-Parse. Read live from the Hugging Face API on November 6, 2026.
| # | Model | Organization | Score |
|---|---|---|---|
| 1 | KDLAI/KDL-Frontier-Parser-nano-v2 | KDLAI | 78.96 |
| 2 | nvidia/Nemotron-3.5-Parse | nvidia | 75.2 |
| 3 | lightonai/LightOnOCR-3-4B | lightonai | 75.1 |
| 4 | lightonai/LightOnOCR-3-0.8B | lightonai | 74.6 |
| 5 | infly/Infinity-Parser2-Pro | infly | 74.3 |
| 6 | infly/Infinity-Parser2-Flash | infly | 73.25 |
| 7 | opendatalab/MinerU2.5-Pro-2605-1.2B | opendatalab | 72.78 |
| 8 | lightonai/LightOnOCR-3-1B | lightonai | 71.4 |
| 9 | Qwen/Qwen3.8-27B | Qwen | 70.79 |
| 10 | zai-org/GLM-5.3-Flash | zai-org | 70.75 |
ParseBench is a dataset that Hugging Face marks as an official benchmark. Its leaderboard is built automatically from the .eval_results files that model repositories publish, so entries are mostly self-reported by model authors and evaluation settings differ between submissions. A verified tag appears only where Hugging Face marks an entry as verified.
Original board: huggingface.co/datasets/llamaindex/ParseBench
Part of the Hugging Face official benchmarks map by QuantID. Unofficial. Not affiliated with or endorsed by Hugging Face.