Benchmark Almanac

OCRBench

GLM-OCR 94% leads since January 2026
MiniCPM-V-2_6 85.2% leads since August 2024
GLM-OCR 94% leads since January 2026
In use
Version 1

The official leaderboard hasn’t scored a model released after Granite Vision 3.3 2B (June 2025). Labs report their own numbers for 30 newer models, such as GLM-OCR at 94%.

Models answer 1,000 checked questions on text recognition, scene text, documents, key information extraction, and handwritten math. The score counts correct answers.

OCRBench v2, released in December 2024 with 10,000 pairs, is a separate version.

Released
May 2023
Kind
Documents, charts and text in images
Level
Everyday
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images, Documents → Text
Interaction
Single turn
Run by
Anyone
In views
Vision
Strongest model tested
Kimi K2.5 #76 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 19th, 2026

In the authors’ words

Large models have recently played a dominant role in natural language processing and multimodal vision-language learning. However, their effectiveness in text-related visual tasks remains relatively unexplored. In this paper, we conducted a comprehensive evaluation of Large Multimodal Models, such as GPT4V and Gemini, in various text-related visual tasks including Text Recognition, Scene Text-Centric Visual Question Answering (VQA), Document-Oriented VQA, Key Information Extraction (KIE), and Handwritten Mathematical Expression Recognition (HMER). To facilitate the assessment of Optical Character Recognition (OCR) capabilities in Large Multimodal Models, we propose OCRBench, a comprehensive evaluation benchmark. OCRBench contains 29 datasets, making it the most comprehensive OCR evaluation benchmark available. Furthermore, our study reveals both the strengths and weaknesses of these models, particularly in handling multilingual text, handwritten text, non-semantic text, and mathematical expression recognition. Most importantly, the baseline results presented in this study could provide a foundational framework for the conception and assessment of innovative strategies targeted at enhancing zero-shot multimodal techniques. The evaluation pipeline and benchmark are available at https://github.com/Yuliang-Liu/MultimodalOCR.

Abstract of OCRBench: On the Hidden Mystery of OCR in Large Multimodal Models (arXiv:2305.07895, CC0)
35
Models scored in lab cards
35
Launch posts list it, from 14 labs
20
System cards report it, from 12 labs

GLM-OCR leads with 94%

Best score per model as labs report them in their own cards, drawn striped. Labs pick their own settings, so these compare less cleanly than independent runs.

MiniCPM-V-2_6 leads with 85.2% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Dotted bars are reruns by Epoch AI or another board, and each row names its source.

GLM-OCR leads with 94%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

  1. 1 GLM-OCR 94%
  2. 2 Qwen3.5 397B-A17B 93.1%
  3. 3 Kimi K2.5 92.3%
  4. 4 Qwen3.5-122B-A10B 92.1%
  5. 5 Qwen3.5-Omni-Plus 91.3%
  6. 6 MiniCPM-V-4.7-35B-A3B 91.2%
  7. 7 Qwen3.5-35B-A3B 91%
  8. 8 InternVL3.5-241B-A28B 90.7%
  9. 9 Qwen3.5-27B 89.4%
  10. 10 Qwen3.6 27B 89.4%
  11. 11 Qwen3.5-9B 89.2%
  12. 12 Qwen2.5-VL-72B 88.5%
  13. 13 ERNIE 5.0 87.8%
  14. 14 Qwen3-VL-235B-A22B 87.5%
  15. 15 Kimi-VL-A3B-Thinking-2506 86.9%
  16. 16 Command A Vision 86.9%
  17. 17 STEP3-VL-10B 86.8%
  18. 18 Nemotron 3 Nano Omni 30B-A3B 86.6%
  19. 19 GLM-4.5V 86.5%
  20. 20 GLM-4.6V 86.5%
  21. 21 Seed1.5-VL 86.1%
  22. 22 ERNIE-4.5-VL-28B-A3B-Thinking 85.8%
  23. 23 Keye-VL-2.0-30B-A3B 85.7%
  24. 24 Nemotron Nano 12B v2 VL 85.6%
  25. 25 Qwen3.5-4B 85%
  26. 26 LongCat-Flash-Omni 84.9%
  27. 27 Qwen3.5-2B 84.5%
  28. 28 Phi-4-multimodal 84.4%
  29. 29 GLM-4.1V-9B-Thinking 84.2%
  30. 30 LFM2.5-VL-3B 84.2%
  31. 31 Llama-3.1-Nemotron-Nano-VL-8B-V1 83.9%
  32. 32 North Micro Vision Instruct 79.2%
  33. 33 Phi-4-Reasoning-Vision-15B 76%
  34. 34 Qwen3.5-0.8B 74.5%
  35. 35 LFM2-VL-1.6B 74.2%

Source: Labs’ model cards

Sources: Labs’ model cards and official OCRBench leaderboard

#ModelScoreSettingDateSource
1 GLM-OCR 94% Jan 2026 Lab model card
2 Qwen3.5 397B-A17B 93.1% Feb 2026 Lab model card
3 Kimi K2.5 92.3% Jan 2026 Lab model card
4 Qwen3.5-122B-A10B 92.1% Feb 2026 Lab model card
5 Qwen3.5-Omni-Plus 91.3% Mar 2026 Lab model card
6 MiniCPM-V-4.7-35B-A3B 91.2% Oct 2026 Lab model card
7 Qwen3.5-35B-A3B 91% Feb 2026 Lab model card
8 InternVL3.5-241B-A28B 90.7% Aug 2025 Lab model card
9 Qwen3.5-27B 89.4% Feb 2026 Lab model card
10 Qwen3.6 27B 89.4% Apr 2026 Lab model card
11 Qwen3.5-9B 89.2% Mar 2026 Lab model card
12 Qwen2.5-VL-72B 88.5% Jan 2025 Lab model card
13 ERNIE 5.0 87.8% Feb 2026 Lab model card
14 Qwen3-VL-235B-A22B 87.5% Sep 2025 Lab model card
15 Kimi-VL-A3B-Thinking-2506 86.9% Jun 2025 Lab model card
16 Command A Vision 86.9% Jul 2025 Lab model card
17 STEP3-VL-10B 86.8% Jan 2026 Lab model card
18 Nemotron 3 Nano Omni 30B-A3B 86.6% Apr 2026 Lab model card
19 GLM-4.5V 86.5% Aug 2025 Lab model card
20 GLM-4.6V 86.5% Dec 2025 Lab model card
21 Seed1.5-VL 86.1% May 2025 Lab model card
22 ERNIE-4.5-VL-28B-A3B-Thinking 85.8% Nov 2025 Lab model card
23 Keye-VL-2.0-30B-A3B 85.7% May 2026 Lab model card
24 Nemotron Nano 12B v2 VL 85.6% Oct 2025 Lab model card
25 Qwen3.5-4B 85% Mar 2026 Lab model card
26 LongCat-Flash-Omni 84.9% Oct 2025 Lab model card
27 Qwen3.5-2B 84.5% Mar 2026 Lab model card
28 Phi-4-multimodal 84.4% Feb 2025 Lab model card
29 GLM-4.1V-9B-Thinking 84.2% Jul 2025 Lab model card
30 LFM2.5-VL-3B 84.2% Aug 2026 Lab model card
31 Llama-3.1-Nemotron-Nano-VL-8B-V1 83.9% Jun 2025 Lab model card
32 North Micro Vision Instruct 79.2% Aug 2026 Lab model card
33 Phi-4-Reasoning-Vision-15B 76% Mar 2026 Lab model card
34 Qwen3.5-0.8B 74.5% Mar 2026 Lab model card
35 LFM2-VL-1.6B 74.2% Aug 2025 Lab model card
#ModelScoreSettingDateSource
1 MiniCPM-V-2_6 85.2% Aug 2024 Official leaderboard
2 Granite Vision 3.3 2B 82.4% Jun 2025 Official leaderboard
3 InternVL2-1B 77.9% Jul 2024 Official leaderboard
4 InternVL2-4B 77.6% Jul 2024 Official leaderboard
5 InternVL2-2B 76.8% Jul 2024 Official leaderboard
6 Qwen-VL-Max 72.3% Jan 2024 Official leaderboard
7 Qwen-VL-Plus 69.4% Jan 2024 Official leaderboard
8 MiniCPM-V 2.0 60.5% Apr 2024 Official leaderboard
9 Monkey 51.4% Nov 2023 Official leaderboard
10 mPLUG-Owl2 36.6% Nov 2023 Official leaderboard
11 LLaVA-v1.5-13B 33.1% Oct 2023 Official leaderboard
#ModelScoreSettingDateSource
1 GLM-OCR 94% Jan 2026 Lab model card
2 Qwen3.5 397B-A17B 93.1% Feb 2026 Lab model card
3 Kimi K2.5 92.3% Jan 2026 Lab model card
4 Qwen3.5-122B-A10B 92.1% Feb 2026 Lab model card
5 Qwen3.5-Omni-Plus 91.3% Mar 2026 Lab model card
6 MiniCPM-V-4.7-35B-A3B 91.2% Oct 2026 Lab model card
7 Qwen3.5-35B-A3B 91% Feb 2026 Lab model card
8 InternVL3.5-241B-A28B 90.7% Aug 2025 Lab model card
9 Qwen3.5-27B 89.4% Feb 2026 Lab model card
10 Qwen3.6 27B 89.4% Apr 2026 Lab model card
11 Qwen3.5-9B 89.2% Mar 2026 Lab model card
12 Qwen2.5-VL-72B 88.5% Jan 2025 Lab model card
13 ERNIE 5.0 87.8% Feb 2026 Lab model card
14 Qwen3-VL-235B-A22B 87.5% Sep 2025 Lab model card
15 Kimi-VL-A3B-Thinking-2506 86.9% Jun 2025 Lab model card
16 Command A Vision 86.9% Jul 2025 Lab model card
17 STEP3-VL-10B 86.8% Jan 2026 Lab model card
18 Nemotron 3 Nano Omni 30B-A3B 86.6% Apr 2026 Lab model card
19 GLM-4.5V 86.5% Aug 2025 Lab model card
20 GLM-4.6V 86.5% Dec 2025 Lab model card
21 Seed1.5-VL 86.1% May 2025 Lab model card
22 ERNIE-4.5-VL-28B-A3B-Thinking 85.8% Nov 2025 Lab model card
23 Keye-VL-2.0-30B-A3B 85.7% May 2026 Lab model card
24 Nemotron Nano 12B v2 VL 85.6% Oct 2025 Lab model card
25 MiniCPM-V-2_6 85.2% Aug 2024 Official leaderboard
26 Qwen3.5-4B 85% Mar 2026 Lab model card
27 LongCat-Flash-Omni 84.9% Oct 2025 Lab model card
28 Qwen3.5-2B 84.5% Mar 2026 Lab model card
29 Phi-4-multimodal 84.4% Feb 2025 Lab model card
30 GLM-4.1V-9B-Thinking 84.2% Jul 2025 Lab model card
31 LFM2.5-VL-3B 84.2% Aug 2026 Lab model card
32 Llama-3.1-Nemotron-Nano-VL-8B-V1 83.9% Jun 2025 Lab model card
33 Granite Vision 3.3 2B 82.4% Jun 2025 Official leaderboard
34 North Micro Vision Instruct 79.2% Aug 2026 Lab model card
35 InternVL2-1B 77.9% Jul 2024 Official leaderboard
36 InternVL2-4B 77.6% Jul 2024 Official leaderboard
37 InternVL2-2B 76.8% Jul 2024 Official leaderboard
38 Phi-4-Reasoning-Vision-15B 76% Mar 2026 Lab model card
39 Qwen3.5-0.8B 74.5% Mar 2026 Lab model card
40 LFM2-VL-1.6B 74.2% Aug 2025 Lab model card
41 Qwen-VL-Max 72.3% Jan 2024 Official leaderboard
42 Qwen-VL-Plus 69.4% Jan 2024 Official leaderboard
43 MiniCPM-V 2.0 60.5% Apr 2024 Official leaderboard
44 Monkey 51.4% Nov 2023 Official leaderboard
45 mPLUG-Owl2 36.6% Nov 2023 Official leaderboard
46 LLaVA-v1.5-13B 33.1% Oct 2023 Official leaderboard

The benchmark's own leaderboard has 29 entries, 11 of them models we track. Switch the scores above to rank them together.

4 models have held number one in labs’ own numbers since its release in May 2023 6 models have held number one across all independent runs since its release in May 2023 10 models have held number one with claims included since its release in May 2023

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

University of Wisconsin-Madison Alibaba Group Huazhong University of Science and Technology Other labs Shanghai AI Lab Qwen Moonshot AI Z.ai
Own leaderboard Passed 85% in Aug 2024, by MiniCPM-V-2_6
Lab claims Passed 85% in Jan 2025, by Qwen2.5-VL-72B
GLM-OCR 94%

Sources: Official OCRBench leaderboard and labs’ model cards

MiniCPM-V-2_6 on the official leaderboard passed 85% in August 2024

That was 15 months after its release. By the same age, 20% of the benchmarks from 2023 had passed 85%. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GLM-OCR Qwen2.5-VL-72B InternVL2-1B Qwen-VL-Max Monkey mPLUG-Owl2 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GLM-OCR MiniCPM-V-2_6

Sources: Official OCRBench leaderboard and labs’ model cards

3 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Qwen
Microsoft
Baidu
ByteDance Seed
NVIDIA
Moonshot AI
Z.ai
Cohere
Liquid AI
Shanghai AI Lab
Meituan
StepFun
Kuaishou
OpenBMB (ModelBest)
DateLaunch postScoreAs the post writes it
Oct 7, 2026 MiniCPM-V-4.7-35B-A3B 91.2 OCRBench
Aug 12, 2026 LFM2.5-VL-3B 84.2 OCRBench v1
Aug 12, 2026 North Micro Vision Instruct 79.2 OCRBench (0.792)
May 25, 2026 Keye-VL-2.0-30B-A3B 85.7 OCRBench
Apr 22, 2026 Qwen3.6 27B 89.4 OCRBench
Mar 30, 2026 Qwen3.5-Omni-Plus 91.3 OCRBench
Mar 4, 2026 Phi-4-Reasoning-Vision-15B 76 OCRBench (default mode)
Mar 2, 2026 Qwen3.5-0.8B 74.5 OCRBench 74.5/79.1 (with CI / without CI)
Mar 2, 2026 Qwen3.5-2B 84.5 OCRBench 84.5/85.4 (with CI / without CI)
Mar 2, 2026 Qwen3.5-4B 85 OCRBench
Mar 2, 2026 Qwen3.5-9B 89.2 OCRBench
Feb 24, 2026 Qwen3.5-35B-A3B 91 OCRBench
Feb 24, 2026 Qwen3.5-27B 89.4 OCRBench
Feb 24, 2026 Qwen3.5-122B-A10B 92.1 OCRBench
Feb 16, 2026 Qwen3.5 397B-A17B 93.1 OCRBench
Feb 6, 2026 ERNIE 5.0 87.8 OCRBench 878 (out of 1000)
Jan 30, 2026 GLM-OCR 94 OCRBench (Text)
Jan 27, 2026 Kimi K2.5 92.3 OCRBench
Jan 13, 2026 STEP3-VL-10B 86.8 OCRBench
Dec 8, 2025 GLM-4.6V 86.5 OCRBench
Nov 11, 2025 ERNIE-4.5-VL-28B-A3B-Thinking 85.8 OCRBench 858 (out of 1000)
Oct 31, 2025 LongCat-Flash-Omni 84.9 OCRBench
Oct 28, 2025 Nemotron Nano 12B v2 VL 85.6 OCRBench
Sep 23, 2025 Qwen3-VL-235B-A22B 87.5 OCRBench 875.0 (out of 1000)
Aug 25, 2025 InternVL3.5-241B-A28B 90.7 OCRBench
Aug 12, 2025 LFM2-VL-1.6B 74.2 OCRBench 742 of 1000
Aug 11, 2025 GLM-4.5V 86.5 OCRBench
Jul 31, 2025 Command A Vision 86.9 OCRBench
Jul 1, 2025 GLM-4.1V-9B-Thinking 84.2 OCRBench
Jun 21, 2025 Kimi-VL-A3B-Thinking-2506 86.9 OCRBench (Acc) 869
Jun 3, 2025 Llama-3.1-Nemotron-Nano-VL-8B-V1 83.9 OCRBench 839 (out of 1000)
May 12, 2025 Seed1.5-VL 86.1 OCRBench
Mar 16, 2025 ERNIE 4.5 OCRBench
Feb 26, 2025 Phi-4-multimodal 84.4 OCR Bench
Jan 28, 2025 Qwen2.5-VL-72B 88.5 OCRBench 885 (out of 1000)

20 system cards report it, from 12 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Jun 9, 2026 Keye-VL-2.0-30B-A3B technical report 85.7 OCRBench (Figure 5, printed value) Capabilities
Apr 27, 2026 Nemotron 3 Nano Omni 30B-A3B technical report 86.6 OCRBench, reasoning on, 88.3 reasoning off (Table 7) Capabilities
Apr 17, 2026 Qwen3.5-Omni-Plus technical report 91.3 OCRBench (Table 6) Capabilities
Mar 11, 2026 GLM-OCR technical report OCRBench (Text), text recognition part only, 94.0, no full OCRBench score (Table 3) Capabilities
Mar 4, 2026 Phi-4-Reasoning-Vision-15B technical report 76 OCRBench, default mode, 75.6 force nothink, 73.7 force thinking (Tables 5 and 6) Capabilities
Feb 4, 2026 ERNIE 5.0 technical report 87.8 OCRBench 878 out of 1000 (Table 3) Capabilities
Jan 30, 2026 Kimi K2.5 technical report 92.3 OCRBench (Table 4) Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 86.8 OCRBench, SeRe (Tables 1 and 3). PaCoRe 89.00 Capabilities
Jan 1, 2026 GLM-4.6V technical report 86.5 OCRBench, GLM-4.6V thinking (Table 2) Capabilities
Nov 28, 2025 LFM2-VL-1.6B technical report 72.9 OCRBench 729 of 1000, LFM2-VL-1.6B (Table 8) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 87.5 OCRBench 875 of 1000, recorded as 87.5% (Table 2) Capabilities
Nov 6, 2025 Nemotron Nano 12B v2 VL technical report 83.5 OCRBench, reasoning on (Table 3). Reasoning off 85.6 Capabilities
Oct 31, 2025 LongCat-Flash-Omni technical report 84.9 OCRBench (Table 5) Capabilities
Aug 25, 2025 InternVL3.5-241B-A28B technical report 90.7 OCRBench 907 out of 1000 (Table 2, Table 4) Capabilities
Aug 13, 2025 GLM-4.5V technical report 86.5 OCRBench, GLM-4.5V thinking (Table 2) Capabilities
Jul 1, 2025 GLM-4.1V-9B-Thinking technical report 84.2 OCRBench, GLM-4.1V-9B-Thinking thinking (Table 2) Capabilities
Jun 23, 2025 Kimi-VL-A3B-Thinking-2506 technical report 86.9 OCRBench (Acc) 869 out of 1000 (Table 5) Capabilities
May 11, 2025 Seed1.5-VL technical report 86.1 OCRBench 861 of 1000, thinking, recorded as 86.1% (Table 6). Non-thinking 881 Capabilities
Mar 3, 2025 Phi-4-multimodal technical report 84.4 OCRBench (Table 1) Capabilities
Feb 19, 2025 Qwen2.5-VL-72B technical report 88.5 OCRBench 885 of 1000, recorded as 88.5% (Table 5) Capabilities

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Qwen Shanghai AI Lab Moonshot AI Z.ai NVIDIA Google DeepMind
OCRBench v2 10 leaders
Qwen3.6 27B 68.3%
OCRBench Passed 85% in Jan 2025, by Qwen2.5-VL-72B
GLM-OCR 94%
OCRBench v2 6 leaders
Qwen3.6 27B 68.3%

Sources: Official leaderboards and labs’ model cards

More in Vision

All 636 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 (29 entries), launch posts (34 scores) and system cards (19 scores).
  • Pass Models scored 46 models we track, plus 18 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for MiniCPM-V-4.7-35B-A3B.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Kimi K2.5 (#76).
  • Note Headroom Saturated. The best is 94%, past 85% since August 2024.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 11 rows for models we track and 18 other entries at the last read.
  • Pass Last read November 19th, 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.

How we check