BHL OCR Eval
Models transcribe scanned pages from historical natural history books.
The board compares character error rates against hand transcriptions and reports separate scores for normalized reading text and original spelling.
- Made by
- finebooks
- Released
- August 2026
- Kind
- Documents, charts and text in images
- Level
- School
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
-
Qwen3.5-9B #128 on the capability index
- Last model added
- October 2026
- Scores refreshed
- November 20th, 2026
dots.ocr tops the leaderboard with 2.35% Unlimited OCR 2 leads with 3.29%
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Unlimited OCR 2, at 3.29%. This view shows only models we track, each at its best.
- 1 dots.ocr 2.35%
- 2 dots.mocr 2.37%
- 3 Chandra-OCR-2 2.85%
- 4 OvisOCR2 3.05%
- 5 LightOnOCR-3-4B 3.12%
- 6
Unlimited OCR 2 3.29%
- 7 kraken PP-OCRv6 3.38%
- 8
MiniCPM-OCR-1.0 3.68%
- 9 OvisOCR3 3.82%
- 10 PaddleOCR-VL-1.6 3.92%
- 11 Surya OCR 2.5 4.32%
- 12 olmOCR-2 4.32%
- 13 LightOnOCR-3-0.8B 4.55%
- 14 LightOnOCR-2 4.89%
- 15 GLM-OCR 4.91%
- 16
Qwen3.5-9B 5.07%
- 17 Qianfan-OCR 5.75%
- 18 Unlimited-OCR 6.04%
- 19 Nemotron-3.5-Parse 6.11%
- 20
DeepSeek-OCR 6.17%
- 21
DeepSeek-OCR 2 6.20%
- 22 Indra OCR 6.42%
- 23 Tesseract 5 6.42%
- 24 SmolDocling 6.60%
- 25 OXR-1.0 11.2%
- 26 Falcon-OCR 14.3%
Source: Official BHL OCR Eval leaderboard
- 1
Unlimited OCR 2 3.29%
- 2
MiniCPM-OCR-1.0 3.68%
- 3 Surya OCR 2.5 4.32%
- 4 GLM-OCR 4.91%
- 5
Qwen3.5-9B 5.07%
- 6
DeepSeek-OCR 6.17%
- 7
DeepSeek-OCR 2 6.20%
- 8 OXR-1.0 11.2%
Source: Official BHL OCR Eval leaderboard
| # | Entry | Score ± | Setting | Date | Link | Source |
|---|---|---|---|---|---|---|
| 1 | dots.ocr 3B | 2.35% 1.6 to 3.2 | huggingface.co | Official leaderboard | ||
| 2 | dots.mocr 3B | 2.37% ±1 | huggingface.co | Official leaderboard | ||
| 3 | Chandra-OCR-2 5.3B | 2.85% 2 to 3.8 | huggingface.co | Official leaderboard | ||
| 4 | OvisOCR2 0.9B | 3.05% 2.3 to 3.6 | huggingface.co | Official leaderboard | ||
| 5 | LightOnOCR-3-4B 4.5B | 3.12% 2.4 to 4 | huggingface.co | Official leaderboard | ||
| 6 | | 3.29% 2.6 to 4.2 | Oct 2026 | huggingface.co | Official leaderboard | |
| 7 | kraken PP-OCRv6 0B | 3.38% ±1 | Official leaderboard | |||
| 8 | | 3.68% 1.5 to 5.4 | Oct 2026 | huggingface.co | Official leaderboard | |
| 9 | OvisOCR3 0.9B | 3.82% ±1 | huggingface.co | Official leaderboard | ||
| 10 | PaddleOCR-VL-1.6 1B | 3.92% 2.6 to 5 | huggingface.co | Official leaderboard | ||
| 11 | Surya OCR 2.5 | 4.32% 3.5 to 5.1 | huggingface.co | Official leaderboard | ||
| 12 | olmOCR-2 8.3B | 4.32% 2 to 6.9 | huggingface.co | Official leaderboard | ||
| 13 | LightOnOCR-3-0.8B 0.9B | 4.55% 3.1 to 5.7 | huggingface.co | Official leaderboard | ||
| 14 | LightOnOCR-2 1B | 4.89% ±2.2 | huggingface.co | Official leaderboard | ||
| 15 | GLM-OCR | 4.91% ±2.3 | Jan 2026 | huggingface.co | Official leaderboard | |
| 16 | | 5.07% 2.9 to 7.7 | Feb 2026 | huggingface.co | Official leaderboard | |
| 17 | Qianfan-OCR 4.7B | 5.75% 3.1 to 8.8 | huggingface.co | Official leaderboard | ||
| 18 | Unlimited-OCR 3.3B | 6.04% 3.5 to 8.2 | huggingface.co | Official leaderboard | ||
| 19 | Nemotron-3.5-Parse 0.6B | 6.11% 4.9 to 7.6 | huggingface.co | Official leaderboard | ||
| 20 | | 6.17% 4.7 to 7.5 | Oct 2025 | huggingface.co | Official leaderboard | |
| 21 | | 6.20% 4.5 to 8.3 | Jan 2026 | huggingface.co | Official leaderboard | |
| 22 | Indra OCR 0B | 6.42% 2.6 to 10.1 | huggingface.co | Official leaderboard | ||
| 23 | Tesseract 5 | 6.42% 4.2 to 7.9 | Official leaderboard | |||
| 24 | SmolDocling 0.3B | 6.60% 3.4 to 10.1 | huggingface.co | Official leaderboard | ||
| 25 | OXR-1.0 | 11.2% 3.3 to 17.9 | huggingface.co | Official leaderboard | ||
| 26 | Falcon-OCR 0.3B | 14.3% 8.9 to 19.3 | huggingface.co | Official leaderboard |
| # | Model | Score ± | Setting | Date | Link | Source |
|---|---|---|---|---|---|---|
| 1 | | 3.29% 2.6 to 4.2 | Oct 2026 | huggingface.co | Official leaderboard | |
| 2 | | 3.68% 1.5 to 5.4 | Oct 2026 | huggingface.co | Official leaderboard | |
| 3 | Surya OCR 2.5 | 4.32% 3.5 to 5.1 | huggingface.co | Official leaderboard | ||
| 4 | GLM-OCR | 4.91% ±2.3 | Jan 2026 | huggingface.co | Official leaderboard | |
| 5 | | 5.07% 2.9 to 7.7 | Feb 2026 | huggingface.co | Official leaderboard | |
| 6 | | 6.17% 4.7 to 7.5 | Oct 2025 | huggingface.co | Official leaderboard | |
| 7 | | 6.20% 4.5 to 8.3 | Jan 2026 | huggingface.co | Official leaderboard | |
| 8 | OXR-1.0 | 11.2% 3.3 to 17.9 | huggingface.co | Official leaderboard |
Rankings by part
We read 6 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Content CER, Sparse CER and de. Unlimited OCR 2 leads 3 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Content CER
8 models we track have a score here.
- 1 GLM-OCR ↑3 2.36%
- 2
Unlimited OCR 2 ↓1 3.11%
- 3
Qwen3.5-9B ↑2 3.14%
- 4
MiniCPM-OCR-1.0 ↓2 3.47%
- 5 Surya OCR 2.5 ↓2 4.08%
- 6
DeepSeek-OCR 5.90%
- 7
DeepSeek-OCR 2 6.00%
- 8 OXR-1.0 11%
Sparse CER
8 models we track have a score here.
- 1
Unlimited OCR 2 190.4%
- 2
DeepSeek-OCR 2 ↑5 209.3%
- 3 OXR-1.0 ↑5 210.7%
- 4
MiniCPM-OCR-1.0 ↓2 228.9%
- 5 Surya OCR 2.5 ↓2 262.4%
- 6
DeepSeek-OCR 296.3%
- 7
Qwen3.5-9B ↓2 1998.3%
- 8 GLM-OCR ↓4 2692.9%
de
8 models we track have a score here.
- 1
Qwen3.5-9B ↑4 3.77%
- 2 Surya OCR 2.5 ↑1 4.72%
- 3
Unlimited OCR 2 ↓2 5.07%
- 4 GLM-OCR 5.35%
- 5
MiniCPM-OCR-1.0 ↓3 5.38%
- 6
DeepSeek-OCR 9.10%
- 7
DeepSeek-OCR 2 10.7%
- 8 OXR-1.0 13.7%
en
8 models we track have a score here.
- 1
Unlimited OCR 2 2.68%
- 2
MiniCPM-OCR-1.0 4.04%
- 3 Surya OCR 2.5 4.60%
- 4
DeepSeek-OCR 2 ↑3 4.93%
- 5
DeepSeek-OCR ↑1 5.84%
- 6 GLM-OCR ↓2 6.74%
- 7
Qwen3.5-9B ↓2 6.96%
- 8 OXR-1.0 15.4%
fr
8 models we track have a score here.
- 1 GLM-OCR ↑3 1.17%
- 2
MiniCPM-OCR-1.0 1.26%
- 3
Qwen3.5-9B ↑2 2.36%
- 4
Unlimited OCR 2 ↓3 2.42%
- 5 OXR-1.0 ↑3 2.47%
- 6 Surya OCR 2.5 ↓3 3.21%
- 7
DeepSeek-OCR ↓1 3.74%
- 8
DeepSeek-OCR 2 ↓1 3.90%
la
8 models we track have a score here.
- 1
Unlimited OCR 2 6.42%
- 2 OXR-1.0 ↑6 6.62%
- 3 Surya OCR 2.5 6.73%
- 4
MiniCPM-OCR-1.0 ↓2 6.77%
- 5 GLM-OCR ↓1 7.04%
- 6
Qwen3.5-9B ↓1 9.61%
- 7
DeepSeek-OCR ↓1 10.8%
- 8
DeepSeek-OCR 2 ↓1 12.1%
2 models have held number one on the official leaderboard since its release in August 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official BHL OCR Eval leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official BHL OCR Eval leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy.
Qwen 3.6 Plus 86%
- MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
- CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
- MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy.
Seed2.1 Pro 90.7%
- MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy.
Qwen 3.8 Max 97.7%
- MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy.
Qwen 3.8 Max 85.9%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (26 entries).
- Pass Models scored 8 models we track, plus 18 other entries on its leaderboard.
- Pass Newest score The newest is from October 2026, for MiniCPM-OCR-1.0.
- Note Frontier models None of the top 10 has a score. The strongest tested is Qwen3.5-9B (#128).
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 8 rows for models we track and 18 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.