leaderboard/leaderboard.csv

rank,model,category,family,parameters,hf_model_id,avg3,clean_overall,clean_text_edit,clean_formula_cdm,clean_table_teds,clean_reading_order_edit,digital_overall,digital_text_edit,digital_formula_cdm,digital_table_teds,digital_reading_order_edit,real_overall,real_text_edit,real_formula_cdm,real_table_teds,real_reading_order_edit,source_type,source_url,score_provenance,note
1,TeleOCR,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,1.2B,StarDoc-AI/TeleOCR,78.41,86.9,0.111,81.01,91.09,,77.47,0.206,72.59,80.45,,70.85,0.302,65.11,77.66,,author-reported,https://huggingface.co/StarDoc-AI/TeleOCR/blob/a99c4e19b0c83f2e11f45f6a579d7a0ad131e9db/README.md#puredocbench,"{""description"": ""PureDocBench scores from the official TeleOCR model card and technical report Table 3. Per-track Overall values are transcribed as author reported, even where displayed rounded components yield a different recomputed score. Avg3 is derived from the displayed track scores: (86.90 + 77.47 + 70.85) / 3 = 78.406666..., rounded to 78.41; this agrees with the paper abstract."", ""hf_revision"": ""a99c4e19b0c83f2e11f45f6a579d7a0ad131e9db"", ""model_card_url"": ""https://huggingface.co/StarDoc-AI/TeleOCR"", ""paper_url"": ""https://arxiv.org/html/2608.12898v3"", ""aliases"": [""NaviDC-OCR""], ""rename_date"": ""2026-09-10"", ""weights_release_date"": ""2026-08-17"", ""paper_version_date"": ""2026-09-10"", ""verified_at_utc"": ""2026-09-17T02:41:24.593517+00:00"", ""missing_metrics"": [""clean.reading_order_edit"", ""digital.reading_order_edit"", ""real.reading_order_edit""], ""evaluation_disclosure"": ""Technical report v3 Appendix D states that the authors removed invalid repetitive Markdown prediction files for 1 clean page and 5 real-degraded pages, and that missing predictions receive zero. No digital-degraded predictions were removed."", ""evaluation_disclosure_url"": ""https://arxiv.org/html/2608.12898v3#A4"", ""independently_reproduced"": false, ""overall_note"": ""The displayed Clean component metrics would yield 87.00, whereas the authors report 86.90. Preserve the reported Overall without adjusting it or inventing additional precision.""}",Author-reported post-publication result. Formerly NaviDC-OCR; renamed TeleOCR on 2026-09-10. Overall transcribed as author reported; Avg3 derived from displayed track scores. Reading-order metrics were not published. The authors disclose removal of 1 clean and 5 real invalid repetitive prediction files and state that missing predictions receive zero.
2,WeVisDoc-4B,End-to-End Specialists,End-to-End Specialists,4B,tencent/WeVisDoc-4B,75.54,79.81,0.213,71.9,88.81,0.34,77.74,0.241,71.31,85.99,0.361,69.08,0.336,63.14,77.64,0.416,author-reported,https://huggingface.co/tencent/WeVisDoc-4B/blob/1754bfa79ba1a6e24aacc001da2e7bb39b9fbc34/README.md#puredocbench,"{""description"": ""Complete PureDocBench track metrics are transcribed from the author-provided TSV and agree with technical report v1 Table 3. Track Overall scores and Avg3 also match the official GitHub README and Hugging Face model card. Published rounding is preserved."", ""hf_revision"": ""1754bfa79ba1a6e24aacc001da2e7bb39b9fbc34"", ""model_card_url"": ""https://huggingface.co/tencent/WeVisDoc-4B"", ""model_card_source_url"": ""https://huggingface.co/tencent/WeVisDoc-4B/blob/1754bfa79ba1a6e24aacc001da2e7bb39b9fbc34/README.md#puredocbench"", ""github_url"": ""https://github.com/Tencent/WeVisDoc"", ""github_revision"": ""bcbc0eacb976ecb8d69675fe6bcee123f5640777"", ""github_source_url"": ""https://github.com/Tencent/WeVisDoc/blob/bcbc0eacb976ecb8d69675fe6bcee123f5640777/README.md#puredocbench"", ""paper_url"": ""https://arxiv.org/html/2609.20423v1#S4.T3"", ""paper_version_date"": ""2026-09-17"", ""verified_at_utc"": ""2026-09-22T03:44:20.699347+00:00"", ""missing_metrics"": [], ""evaluation_disclosure"": ""Technical report v1 Section 3.4.3 and Table 3 state that each checkpoint is evaluated using the official benchmark protocol and that reported scores are means over three inference runs. The report does not specify an exact ground-truth archive revision, evaluator commit, or evaluated page manifest."", ""evaluation_disclosure_url"": ""https://arxiv.org/html/2609.20423v1#S3.SS4.SSS3"", ""inference_runs"": 3, ""score_aggregation"": ""mean over three inference runs"", ""ground_truth_revision"": null, ""evaluator_revision"": null, ""sample_manifest"": null, ""independently_reproduced"": false, ""detailed_results_url"": ""https://huggingface.co/datasets/zhihengli-casia/puredocbench/blob/main/leaderboard/submissions/wevisdoc/results.tsv""}",Author-reported post-publication result. The technical report states that scores are means over three inference runs. Ground-truth and evaluator revisions are unspecified; these results have not been independently reproduced against gt-latest.
3,OvisOCR2,End-to-End Specialists,End-to-End Specialists,0.8B,ATH-MaaS/OvisOCR2,75.06,81.55,,,,,77.09,,,,,66.56,,,,,author-reported,https://huggingface.co/ATH-MaaS/OvisOCR2/blob/1fc9221b7823a371d6e97f92d527cc847e24e107/performance_puredocbench.png,"{""description"": ""PureDocBench Table 3 image from the official OvisOCR2 model card, visually checked against technical report Table 3. Preserves the authors' published Avg3 of 75.06 (the displayed rounded track values average to 75.0667)."", ""hf_revision"": ""1fc9221b7823a371d6e97f92d527cc847e24e107"", ""model_card_url"": ""https://huggingface.co/ATH-MaaS/OvisOCR2"", ""paper_url"": ""https://arxiv.org/html/2607.13639v1#S4.T3"", ""model_repo_created_at"": ""2026-07-13T05:43:32.000Z"", ""paper_version_date"": ""2026-07-15"", ""verified_at_utc"": ""2026-09-17T02:41:24.593619+00:00"", ""missing_metrics"": [""clean.text_edit"", ""clean.formula_cdm"", ""clean.table_teds"", ""clean.reading_order_edit"", ""digital.text_edit"", ""digital.formula_cdm"", ""digital.table_teds"", ""digital.reading_order_edit"", ""real.text_edit"", ""real.formula_cdm"", ""real.table_teds"", ""real.reading_order_edit""], ""comparison_note"": ""TeleOCR authors also publish an OvisOCR2 comparison row with different rerun scores (82.14/77.77/66.61). This leaderboard entry uses OvisOCR2 authors' own official model card and report scores (81.55/77.09/66.56)."", ""name_note"": ""OvisOCR2 is a separate release from ATH-MaaS/OvisOCR (1.3B). The first-generation OvisOCR official model card was checked and contains no PureDocBench score."", ""independently_reproduced"": false}",Author-reported post-publication result for OvisOCR2. The official model card and technical report publish per-track Overall and Avg3 only; component metrics are unavailable.
4,Qwen3.5-122B-A10B,General VLMs,General VLMs: Qwen3.5,122B/10B,Qwen/Qwen3.5-122B-A10B,74.11,76.14,0.226,67.96,83.03,0.375,76.34,0.22,67.82,83.21,0.366,69.85,0.281,62.19,75.44,0.401,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
5,FD-RL,End-to-End Specialists,End-to-End Specialists,4B,DocTron/FD-RL,73.92,78.38,0.193,68.21,86.22,0.334,76.33,0.214,67.16,83.22,0.35,67.04,0.298,58.82,72.08,0.391,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
6,WeVisDoc-2B,End-to-End Specialists,End-to-End Specialists,2B,tencent/WeVisDoc-2B,73.86,79.36,0.214,71.63,87.79,0.347,76.62,0.247,70.74,83.84,0.37,65.6,0.361,58.9,74.05,0.437,author-reported,https://huggingface.co/tencent/WeVisDoc-2B/blob/86ecb7330566296b695d4f91aea807a260cc84a7/README.md#puredocbench,"{""description"": ""Complete PureDocBench track metrics are transcribed from the author-provided TSV and agree with technical report v1 Table 3. Track Overall scores and Avg3 also match the official GitHub README and Hugging Face model card. Published rounding is preserved."", ""hf_revision"": ""86ecb7330566296b695d4f91aea807a260cc84a7"", ""model_card_url"": ""https://huggingface.co/tencent/WeVisDoc-2B"", ""model_card_source_url"": ""https://huggingface.co/tencent/WeVisDoc-2B/blob/86ecb7330566296b695d4f91aea807a260cc84a7/README.md#puredocbench"", ""github_url"": ""https://github.com/Tencent/WeVisDoc"", ""github_revision"": ""bcbc0eacb976ecb8d69675fe6bcee123f5640777"", ""github_source_url"": ""https://github.com/Tencent/WeVisDoc/blob/bcbc0eacb976ecb8d69675fe6bcee123f5640777/README.md#puredocbench"", ""paper_url"": ""https://arxiv.org/html/2609.20423v1#S4.T3"", ""paper_version_date"": ""2026-09-17"", ""verified_at_utc"": ""2026-09-22T03:44:20.699347+00:00"", ""missing_metrics"": [], ""evaluation_disclosure"": ""Technical report v1 Section 3.4.3 and Table 3 state that each checkpoint is evaluated using the official benchmark protocol and that reported scores are means over three inference runs. The report does not specify an exact ground-truth archive revision, evaluator commit, or evaluated page manifest."", ""evaluation_disclosure_url"": ""https://arxiv.org/html/2609.20423v1#S3.SS4.SSS3"", ""inference_runs"": 3, ""score_aggregation"": ""mean over three inference runs"", ""ground_truth_revision"": null, ""evaluator_revision"": null, ""sample_manifest"": null, ""independently_reproduced"": false, ""detailed_results_url"": ""https://huggingface.co/datasets/zhihengli-casia/puredocbench/blob/main/leaderboard/submissions/wevisdoc/results.tsv""}",Author-reported post-publication result. The technical report states that scores are means over three inference runs. Ground-truth and evaluator revisions are unspecified; these results have not been independently reproduced against gt-latest.
7,Logics-Parsing-v2,End-to-End Specialists,End-to-End Specialists,4B,Logics-MLLM/Logics-Parsing-v2,72.61,76.35,0.213,67.67,82.67,0.342,73.85,0.248,67.33,79.02,0.375,67.64,0.304,61.65,71.64,0.416,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
8,Qwen3.5-9B,General VLMs,General VLMs: Qwen3.5,9B,Qwen/Qwen3.5-9B,70.89,73.87,0.254,67.6,79.39,0.388,73.34,0.26,67.0,79.01,0.396,65.45,0.332,60.91,68.59,0.437,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
9,Gemini-3.1-Pro,General VLMs,General VLMs: Other,---,,70.43,70.04,0.306,65.63,75.08,0.409,69.28,0.322,65.81,74.24,0.417,71.98,0.3,68.62,77.26,0.386,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
10,DotsMOCR,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,3B,dots-studio/dots.mocr,70.39,76.27,0.151,66.23,77.65,0.273,73.16,0.198,64.32,74.95,0.309,61.73,0.312,54.39,61.97,0.393,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
11,Kimi K2.6,General VLMs,General VLMs: Other,1T/32B,moonshotai/Kimi-K2.6,70.1,72.32,0.303,66.93,80.3,0.466,69.95,0.322,64.69,77.31,0.475,68.02,0.335,62.44,75.14,0.481,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
12,MinerU2.5-Pro,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,1.2B,opendatalab/MinerU2.5-Pro-2604-1.2B,70.07,75.87,0.222,65.14,84.68,0.346,71.77,0.272,61.79,80.73,0.378,62.56,0.375,52.7,72.47,0.446,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
13,Qwen3.5-4B,General VLMs,General VLMs: Qwen3.5,4B,Qwen/Qwen3.5-4B,69.82,73.45,0.276,69.96,78.02,0.41,72.53,0.281,68.88,76.78,0.412,63.47,0.38,61.27,67.17,0.477,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
14,Qwen3.5-27B,General VLMs,General VLMs: Qwen3.5,27B,Qwen/Qwen3.5-27B,69.57,72.07,0.227,66.36,72.51,0.362,70.73,0.236,64.61,71.17,0.367,65.92,0.283,61.23,64.82,0.39,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
15,OCRVerse,End-to-End Specialists,End-to-End Specialists,4B,DocTron/OCRVerse,69.4,73.18,0.273,63.78,83.09,0.393,71.36,0.302,63.95,80.36,0.415,63.66,0.363,57.03,70.3,0.452,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
16,Qwen3-VL-8B,General VLMs,General VLMs: Qwen3-VL,8B,Qwen/Qwen3-VL-8B-Instruct,69.07,72.44,0.261,65.1,78.35,0.411,72.03,0.266,64.88,77.82,0.409,62.73,0.342,55.55,66.81,0.448,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
17,YouTu-Parsing,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,2B,tencent/Youtu-Parsing,68.32,75.02,0.23,67.34,80.74,0.358,69.66,0.27,61.44,74.49,0.388,60.29,0.36,52.2,64.69,0.43,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
18,MinerU2.5,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,1.2B,opendatalab/MinerU2.5-2509-1.2B,67.66,74.9,0.184,62.08,81.04,0.327,68.92,0.245,56.99,74.24,0.374,59.15,0.37,49.01,65.41,0.446,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
19,Qwen3-VL-4B,General VLMs,General VLMs: Qwen3-VL,4B,Qwen/Qwen3-VL-4B-Instruct,67.5,72.04,0.262,65.1,77.17,0.418,70.84,0.272,63.54,76.13,0.425,59.61,0.378,55.15,61.47,0.48,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
20,PaddleOCR-VL-1.5,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,0.9B,PaddlePaddle/PaddleOCR-VL-1.5,66.75,73.01,0.266,63.53,82.12,0.428,66.73,0.339,58.03,76.07,0.478,60.5,0.398,54.0,67.33,0.51,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
21,Qwen3.5-397B-A17B,General VLMs,General VLMs: Qwen3.5,397B/17B,Qwen/Qwen3.5-397B-A17B,66.72,69.12,0.233,65.26,65.4,0.366,68.34,0.244,63.91,65.53,0.376,62.7,0.287,60.7,56.12,0.399,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
22,Qwen3.5-35B-A3B,General VLMs,General VLMs: Qwen3.5,35B/3B,Qwen/Qwen3.5-35B-A3B,65.68,68.4,0.232,64.94,63.45,0.374,68.04,0.245,64.78,63.86,0.379,60.59,0.31,59.68,53.07,0.419,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
23,FireRed-OCR,End-to-End Specialists,End-to-End Specialists,2B,FireRedTeam/FireRed-OCR,65.57,70.81,0.287,63.86,77.23,0.396,68.49,0.319,62.64,74.77,0.422,57.42,0.415,51.6,62.16,0.474,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
24,dots.ocr,End-to-End Specialists,End-to-End Specialists,2.9B,dots-studio/dots.ocr,64.55,72.01,0.248,61.37,79.51,0.379,65.95,0.307,56.67,71.86,0.417,55.68,0.403,47.7,59.63,0.467,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
25,olmOCR-2-7B,End-to-End Specialists,End-to-End Specialists,7B,allenai/olmOCR-2-7B-1025,63.78,69.36,0.284,56.89,79.59,0.358,65.87,0.318,54.57,74.81,0.378,56.1,0.417,48.79,61.25,0.439,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
26,GLM-OCR,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,0.9B,zai-org/GLM-OCR,63.34,68.65,0.314,57.89,79.44,0.47,63.06,0.383,53.23,74.21,0.52,58.31,0.433,50.34,67.83,0.543,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
27,Qwen3.5-2B,General VLMs,General VLMs: Qwen3.5,2B,Qwen/Qwen3.5-2B,62.46,66.24,0.348,62.84,70.7,0.473,65.22,0.35,58.3,72.36,0.477,55.92,0.44,50.99,60.79,0.521,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
28,Qwen3-VL-2B,General VLMs,General VLMs: Qwen3-VL,2B,Qwen/Qwen3-VL-2B-Instruct,62.09,66.37,0.3,59.04,70.03,0.439,65.81,0.314,60.25,68.52,0.448,54.09,0.428,51.05,53.99,0.511,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
29,HunyuanOCR,End-to-End Specialists,End-to-End Specialists,1B,tencent/HunyuanOCR,60.56,65.61,0.269,55.74,68.02,0.382,61.49,0.308,51.62,63.68,0.4,54.58,0.421,48.3,57.54,0.459,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
30,Nanonets-OCR2,End-to-End Specialists,End-to-End Specialists,3B,nanonets/Nanonets-OCR2-3B,58.36,64.83,0.254,44.98,74.94,0.377,61.23,0.307,45.4,68.97,0.408,49.03,0.435,35.5,55.09,0.468,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
31,Dolphin-v2,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,3B,ByteDance/Dolphin-v2,57.02,65.9,0.342,59.8,72.12,0.429,60.24,0.393,52.2,67.86,0.461,44.92,0.553,39.98,50.04,0.558,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
32,Qwen3.5-0.8B,General VLMs,General VLMs: Qwen3.5,0.8B,Qwen/Qwen3.5-0.8B,55.99,60.77,0.376,54.39,65.54,0.5,59.28,0.386,54.22,62.22,0.51,47.93,0.498,44.6,48.98,0.557,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
33,olmOCR-7B,End-to-End Specialists,End-to-End Specialists,7B,allenai/olmOCR-7B-0825,55.9,62.56,0.388,58.69,67.77,0.466,57.84,0.436,55.44,61.66,0.499,47.3,0.542,46.26,49.8,0.568,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
34,MonkeyOCR-pro-3B,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,3B,echo840/MonkeyOCR-pro-3B,55.37,62.23,0.346,48.46,72.83,0.492,57.4,0.397,45.57,66.32,0.526,46.49,0.511,38.18,52.43,0.6,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
35,MonkeyOCR-pro-1.2B,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,1.2B,echo840/MonkeyOCR-pro-1.2B,53.54,61.09,0.358,47.43,71.6,0.498,55.72,0.416,43.91,64.83,0.529,43.82,0.556,36.94,50.07,0.609,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
36,OpenDoc-0.1B,End-to-End Specialists,End-to-End Specialists,0.1B,,52.0,60.28,0.411,53.09,68.86,0.519,52.46,0.501,48.41,59.04,0.577,44.27,0.547,38.46,49.06,0.603,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,OpenDoc-0.1B retains the published Avg3 of 52.00. The mean of its displayed track scores is 52.34; this transcription preserves the paper value pending an erratum.
37,Qianfan-OCR,End-to-End Specialists,End-to-End Specialists,4B,baidu/Qianfan-OCR,51.04,57.22,0.37,49.79,58.83,0.443,50.85,0.438,44.41,51.96,0.485,45.06,0.494,39.08,45.53,0.509,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
38,Step3-VL,General VLMs,General VLMs: Other,10B,stepfun-ai/Step3-VL-10B,50.48,53.65,0.496,53.41,57.16,0.509,52.74,0.516,53.62,56.15,0.529,45.06,0.579,45.42,47.66,0.573,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
39,DeepSeek-OCR-2,End-to-End Specialists,End-to-End Specialists,3B,deepseek-ai/DeepSeek-OCR-2,49.51,55.53,0.354,46.0,56.01,0.466,49.41,0.412,40.78,48.67,0.493,43.6,0.486,37.3,42.06,0.533,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
40,UniRec-0.1B,End-to-End Specialists,End-to-End Specialists,0.1B,topdu/unirec_0_1b_onnx,48.59,58.91,0.422,51.31,67.6,0.526,52.42,0.501,48.37,59.04,0.578,34.44,0.658,30.97,38.16,0.685,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
41,DeepSeek-OCR,End-to-End Specialists,End-to-End Specialists,3B,deepseek-ai/DeepSeek-OCR,46.98,53.5,0.419,45.39,57.06,0.514,46.95,0.478,39.99,48.64,0.548,40.48,0.537,34.04,41.12,0.575,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
42,MiniCPM-V-4.5,General VLMs,General VLMs: Other,8B,openbmb/MiniCPM-V-4_5,46.26,51.81,0.439,45.97,53.36,0.481,49.38,0.461,42.79,51.5,0.489,37.59,0.583,32.01,39.06,0.552,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
43,OCRFlux-3B,End-to-End Specialists,End-to-End Specialists,3B,ChatDOC/OCRFlux-3B,42.06,47.14,0.454,38.35,48.46,0.424,41.82,0.486,31.9,42.17,0.437,37.21,0.559,32.65,34.87,0.491,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,
44,OpenOCR,Pipeline / Multi-stage Specialists,Pipeline / Multi-stage Specialists,0.1B,topdu/OpenOCR,29.49,32.7,0.354,33.5,0.0,0.507,30.03,0.41,31.09,0.0,0.541,25.73,0.486,25.81,0.0,0.591,paper,https://arxiv.org/abs/2605.07492,Original PureDocBench paper Table 2,