Benchmark Almanac

OCRBench v2

KDL Frontier (2026-10, EN) 76.9% tops its leaderboard Qwen3.6 27B 68.3% leads the models we track
Qwen 3.8 Max 74.2% leads since August 2026
In use
Version 2

The official leaderboard hasn’t scored a model released after Nemotron 3.5 Super VL (October 2026). Labs report their own numbers for one newer model: MiniCPM-OCR-1.0 at 54.7%.

Visual-text localization and reasoning tasks across 31 scenarios (documents, scenes, charts). Averaged task accuracy.

Released
December 2024
Kind
Documents, charts and text in images
Level
Everyday
Scored
Checked automatically
Answers
Short answer
Tasks
Public, plus a private test set
Measures
Capability
Language
English, Chinese
In and out
Text, Images → Text
Interaction
Single turn
Run by
A host
Based on
OCRBench
In views
Vision
Strongest model tested
Qwen 3.8 Max #30 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 18th, 2026

In the authors’ words

Scoring the Optical Character Recognition (OCR) capabilities of Large Multimodal Models (LMMs) has witnessed growing interest. Existing benchmarks have highlighted the impressive performance of LMMs in text recognition; however, their abilities in certain challenging tasks, such as text localization, handwritten content extraction, and logical reasoning, remain underexplored. To bridge this gap, we introduce OCRBench v2, a large-scale bilingual text-centric benchmark with currently the most comprehensive set of tasks (4x more tasks than the previous multi-scene benchmark OCRBench), the widest coverage of scenarios (31 diverse scenarios), and thorough evaluation metrics, with 10,000 human-verified question-answering pairs and a high proportion of difficult samples. Moreover, we construct a private test set with 1,500 manually annotated images. The consistent evaluation trends observed across both public and private test sets validate the OCRBench v2's reliability. After carefully benchmarking state-of-the-art LMMs, we find that most LMMs score below 50 (100 in total) and suffer from five-type limitations, including less frequently encountered text recognition, fine-grained perception, layout perception, complex element parsing, and logical reasoning. The project website is at: https://99franklin.github.io/ocrbench_v2/

Abstract of OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning (arXiv:2501.00321, CC0)
26
Entries on its own leaderboard, 15 models we track
20
Launch posts list it, from 9 labs
16
System cards report it, from 6 labs

KDL Frontier (2026-10, EN) tops the leaderboard with 76.9% Qwen3.6 27B leads with 68.3%

Best score per entry from the benchmark's own leaderboard, read on November 18th, 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 Qwen3.6 27B, at 68.3%. This view shows only models we track, each at its best.

Qwen 3.8 Max leads with 74.2%, 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 KDL Frontier (2026-10, EN) 76.9%
  2. 2 SEEU DOC 76.2%
  3. 3 Yaochi-DTS-A2T-1.9 73.4%
  4. 4 U2-OCR 71.7%
  5. 5 Posicube_DVLM 69.8%
  6. 6 Qwen3.6 27B 68.3%
  7. 7 KDL Frontier 68.1%
  8. 8 Nemotron 3.5 Super VL 67.4%
  9. 9 Nemotron 3 Nano Omni 30B-A3B 65.8%
  10. 10 Qwen 3.6 35B-A3B 65.5%
  11. 11 Qwen3.5-35B-A3B 65.3%
  12. 12 Qwen 3.8 27B 63.5%
  13. 13 Gemini 3 Pro 63.4%
  14. 14 Seed1.6-vision 62.2%
  15. 15 TeleMM-2.0 61.8%
  16. 16 Qwen3-Omni-30B-A3B 61.3%
  17. 17 Nemotron Nano 12B v2 VL 61.2%
  18. 18 SenseNova-U1.5-8B-MoT 59.9%
  19. 19 Gemini-2.5-Pro (snapshot unspecified) 59.3%
  20. 20 GLM-4.6V-Flash 59%
  21. 21 Qwen3.5-9B 58.7%
  22. 22 SenseNova-U1 58.2%
  23. 23 Muse Glimmer 56.9%
  24. 24 Llama-3.1-Nemotron-Nano-VL-8B-V1 56.4%
  25. 25 GPT-5 55.5%
  26. 26 LFM2.5-VL-3B 42.6%

Source: Official OCRBench v2 leaderboard

Sources: Labs’ model cards and official OCRBench v2 leaderboard

#EntryScoreSettingReleasedRun onSource
1 KDL Frontier (2026-10, EN) 76.9% Sep 2026 Official leaderboard
2 SEEU DOC 76.2% Sep 2026 Official leaderboard
3 Yaochi-DTS-A2T-1.9 73.4% Sep 2026 Official leaderboard
4 U2-OCR 71.7% Sep 2026 Official leaderboard
5 Posicube_DVLM 69.8% Sep 2026 Official leaderboard
6 Qwen3.6 27B 68.3% Apr 2026 Sep 2026 Official leaderboard
7 KDL Frontier 68.1% Sep 2026 Official leaderboard
8 Nemotron 3.5 Super VL 67.4% Oct 2026 Sep 2026 Official leaderboard
9 Nemotron 3 Nano Omni 30B-A3B 65.8% Apr 2026 Sep 2026 Official leaderboard
10 Qwen 3.6 35B-A3B 65.5% Apr 2026 Sep 2026 Official leaderboard
11 Qwen3.5-35B-A3B 65.3% Feb 2026 Sep 2026 Official leaderboard
12 Qwen 3.8 27B 63.5% Aug 2026 Sep 2026 Official leaderboard
13 Gemini 3 Pro 63.4% Nov 2025 Sep 2026 Official leaderboard
14 Seed1.6-vision 62.2% Sep 2026 Official leaderboard
15 TeleMM-2.0 61.8% Sep 2026 Official leaderboard
16 Qwen3-Omni-30B-A3B 61.3% Sep 2025 Sep 2026 Official leaderboard
17 Nemotron Nano 12B v2 VL 61.2% Oct 2025 Sep 2026 Official leaderboard
18 SenseNova-U1.5-8B-MoT 59.9% Sep 2026 Official leaderboard
19 Gemini-2.5-Pro (snapshot unspecified) 59.3% Sep 2026 Official leaderboard
20 GLM-4.6V-Flash 59% Dec 2025 Sep 2026 Official leaderboard
21 Qwen3.5-9B 58.7% Feb 2026 Sep 2026 Official leaderboard
22 SenseNova-U1 58.2% Sep 2026 Official leaderboard
23 Muse Glimmer 56.9% Aug 2026 Sep 2026 Official leaderboard
24 Llama-3.1-Nemotron-Nano-VL-8B-V1 56.4% Jun 2025 Sep 2026 Official leaderboard
25 GPT-5 55.5% Aug 2025 Sep 2026 Official leaderboard
26 LFM2.5-VL-3B 42.6% Aug 2026 Sep 2026 Official leaderboard
#ModelScoreSettingReleasedRun onSource
1 Qwen3.6 27B 68.3% Apr 2026 Sep 2026 Official leaderboard
2 Nemotron 3.5 Super VL 67.4% Oct 2026 Sep 2026 Official leaderboard
3 Nemotron 3 Nano Omni 30B-A3B 65.8% Apr 2026 Sep 2026 Official leaderboard
4 Qwen 3.6 35B-A3B 65.5% Apr 2026 Sep 2026 Official leaderboard
5 Qwen3.5-35B-A3B 65.3% Feb 2026 Sep 2026 Official leaderboard
6 Qwen 3.8 27B 63.5% Aug 2026 Sep 2026 Official leaderboard
7 Gemini 3 Pro 63.4% Nov 2025 Sep 2026 Official leaderboard
8 Qwen3-Omni-30B-A3B 61.3% Sep 2025 Sep 2026 Official leaderboard
9 Nemotron Nano 12B v2 VL 61.2% Oct 2025 Sep 2026 Official leaderboard
10 GLM-4.6V-Flash 59% Dec 2025 Sep 2026 Official leaderboard
11 Qwen3.5-9B 58.7% Feb 2026 Sep 2026 Official leaderboard
12 Muse Glimmer 56.9% Aug 2026 Sep 2026 Official leaderboard
13 Llama-3.1-Nemotron-Nano-VL-8B-V1 56.4% Jun 2025 Sep 2026 Official leaderboard
14 GPT-5 55.5% Aug 2025 Sep 2026 Official leaderboard
15 LFM2.5-VL-3B 42.6% Aug 2026 Sep 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen 3.8 Max 74.2% Aug 2026 Lab model card
2 Qwen3.7-Plus 70.7% Jun 2026 Lab model card
3 MiniCPM-V-4.7-35B-A3B 68.5% Oct 2026 Lab model card
4 Qwen3.6 27B 68.3% Apr 2026 Official leaderboard
5 Nemotron 3.5 Super VL 68.1% Oct 2026 Lab model card
6 Nemotron 3 Nano Omni 30B-A3B 67% Apr 2026 Lab model card
7 Qwen3-VL-235B-A22B 66.8% Sep 2025 Lab model card
8 Qwen 3.6 35B-A3B 65.5% Apr 2026 Official leaderboard
9 Qwen3.5-35B-A3B 65.3% Feb 2026 Official leaderboard
10 GLM-4.6V 65.1% Dec 2025 Lab model card
11 Nova 2.0 Pro Preview 64.5% Dec 2025 Lab model card
12 Qwen 3.8 27B 63.5% Aug 2026 Official leaderboard
13 Gemini 3 Pro 63.4% Nov 2025 Official leaderboard
14 EXAONE 4.5 33B 63.2% Apr 2026 Lab model card
15 Seed2.1 Pro 63.2% Jun 2026 Lab model card
16 Seed2.0 Pro 62.5% Feb 2026 Lab model card
17 Nemotron Nano 12B v2 VL 62% Oct 2025 Lab model card
18 Qwen2.5-VL-72B 61.5% Jan 2025 Lab model card
19 Qwen3-Omni-30B-A3B 61.3% Sep 2025 Official leaderboard
20 Llama-3.1-Nemotron-Nano-VL-8B-V1 60.1% Jun 2025 Lab model card
21 Intern-S1-Pro 60.1% Feb 2026 Lab model card
22 Cosmos3-Nano 60.1% May 2026 Lab model card
23 GLM-4.6V-Flash 59% Dec 2025 Official leaderboard
24 Qwen3.5-9B 58.7% Feb 2026 Official leaderboard
25 Nova 2 Omni Preview 58.2% Dec 2025 Lab model card
26 Qwen2.5-Omni-7B 57.8% Mar 2025 Lab model card
27 Amazon Nova Premier 56.9% Apr 2025 Lab model card
28 Muse Glimmer 56.9% Aug 2026 Official leaderboard
29 Nova 2 Lite 56.1% Dec 2025 Lab model card
30 GPT-5 55.5% Aug 2025 Official leaderboard
31 MiniCPM-OCR-1.0 54.7% Oct 2026 Lab model card
32 LFM2.5-VL-3B 47.5% Aug 2026 Lab model card
33 Cosmos3-Edge 43.7% May 2026 Lab model card
34 North Micro Vision Instruct 36.7% Aug 2026 Lab model card

Labs report scores for 24 models in their cards. Switch the scores above to rank them together.

Rankings by part

We read 9 parts from the benchmark’s own leaderboard in 2 groups: task type and language. Each ranks models on one part of the test. Gemini 3 Pro leads 3 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Recognition

15 models we track have a score here.

  1. 1 Qwen 3.8 27B ↑5 75.5%
  2. 2 GLM-4.6V-Flash ↑8 75.3%
  3. 3 Qwen 3.6 35B-A3B ↑1 74.9%
  4. 4 Nemotron 3.5 Super VL ↓2 74%
  5. 5 Qwen3.5-9B ↑6 72.4%
  6. 6 Qwen3-Omni-30B-A3B ↑2 72.3%
  7. 7 Qwen3.6 27B ↓6 71.1%
  8. 8 Nemotron 3 Nano Omni 30B-A3B ↓5 70.5%
  9. 9 Qwen3.5-35B-A3B ↓4 69.8%
  10. 10 GPT-5 ↑4 69.3%

6 models have held number one on the official leaderboard since its release in December 2024 5 models have held number one with claims included since its release in December 2024

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.

NVIDIA Qwen Google DeepMind
Own leaderboard 6 leaders
Qwen3.6 27B 68.3%
Lab claims 5 leaders
Qwen 3.8 Max 74.2%

Sources: Official OCRBench v2 leaderboard and labs’ model cards

Scores over time

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% 2025 2026 2027 Near the ceiling (85%) Qwen 3.8 Max Qwen3.6 27B Qwen3-VL-235B-A22B Qwen2.5-VL-72B 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Qwen 3.8 Max Qwen3-VL-235B-A22B

Sources: Official OCRBench v2 leaderboard and labs’ model cards

5 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
Amazon
NVIDIA
Z.ai
Shanghai AI Lab
LG AI Research
Cohere
Liquid AI
OpenBMB (ModelBest)
DateLaunch postScoreAs the post writes it
Oct 28, 2026 MiniCPM-OCR-1.0 54.7 OCRBench v2
Oct 15, 2026 Nemotron 3.5 Super VL 68.1 OCRBench v2, EN 68.12 / ZH 59.15
Oct 7, 2026 MiniCPM-V-4.7-35B-A3B 68.5 OCRBench_v2
Aug 12, 2026 LFM2.5-VL-3B 47.5 OCRBench v2 (En)
Aug 12, 2026 North Micro Vision Instruct 36.7 OCRBench v2_en
Aug 3, 2026 Qwen 3.8 Max 74.2 OCR-Bench-V2 (EN/ZH) 74.2 / 68.3
Jun 1, 2026 Qwen3.7-Plus 70.7 OCR-Bench-V2(EN) 70.7, OCR-Bench-V2(ZH) 67.1
Apr 28, 2026 Nemotron 3 Nano Omni 30B-A3B 67 OCRBenchV2 (EN)
Apr 9, 2026 EXAONE 4.5 33B 63.2 OCRBench v2
Feb 2, 2026 Intern-S1-Pro 60.1 OCRBench V2 (ENG / CHN) 60.1 / 60.6
Dec 8, 2025 GLM-4.6V 65.1 OCR-Bench_v2 (EN)
Dec 2, 2025 Nova 2 Omni Preview 58.2 OCRBench-v2 avg accuracy
Dec 2, 2025 Nova 2 Lite 56.1 OCRBench-v2 avg accuracy
Dec 2, 2025 Nova 2.0 Pro Preview 64.5 OCRBench-v2 avg accuracy
Oct 28, 2025 Nemotron Nano 12B v2 VL 62 OCRBenchv2 (English)
Sep 23, 2025 Qwen3-VL-235B-A22B 66.8 OCRBenchV2 (en/ch) 66.8 / 63.5
Jun 3, 2025 Llama-3.1-Nemotron-Nano-VL-8B-V1 60.1 OCRBenchV2 English 60.1 / Chinese 37.9
Apr 30, 2025 Amazon Nova Premier 56.9 OCRBench-v2
Mar 27, 2025 Qwen2.5-Omni-7B 57.8 OCRBench_V2en
Jan 28, 2025 Qwen2.5-VL-72B 61.5 OCRBench-V2(en/zh) 61.5/63.7

16 system cards report it, from 6 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
Oct 15, 2026 Nemotron 3.5 Super VL technical report 68.1 OCRBench-v2 (EN) 68.12, (ZH) 59.15 (Table 10) Capabilities
Jun 23, 2026 Seed2.1 Pro model card 63.2 OCRBenchv2 (Table 9) Capabilities
May 31, 2026 Cosmos3-Nano technical report 60.1 OCRBench-v2, Cosmos3-Nano 8B column, Table 10 Capabilities
May 31, 2026 Cosmos3-Edge technical report 43.7 OCRBench-v2, Cosmos3-Edge 2B column, Table 10 Capabilities
Apr 27, 2026 Nemotron 3 Nano Omni 30B-A3B technical report 67 OCRBenchV2 (EN), reasoning on, ZH 52.7. Reasoning off EN 65.8, ZH 52.0 (Table 7) Capabilities
Apr 9, 2026 EXAONE 4.5 33B technical report 63.2 OCRBench v2 (Table 2, Reasoning mode) Capabilities
Mar 26, 2026 Intern-S1-Pro technical report 60.1 OCRBench V2 (ENG / CHN) 60.1 / 60.6, English recorded, non-thinking (Table 2) Capabilities
Feb 14, 2026 Seed2.0 Pro model card 62.5 OCRBenchv2, mean of English and Chinese overall (Table 8) Capabilities
Dec 2, 2025 Nova 2 Omni Preview technical report 58.2 OCRBench-v2 avg accuracy (Table 6). The 1.2 text says 58.5 Capabilities
Dec 2, 2025 Nova 2 Lite technical report 56.1 OCRBench-v2 avg accuracy (Table 3) Capabilities
Dec 2, 2025 Nova 2.0 Pro Preview technical report 64.5 OCRBench-v2 avg accuracy (Table 3) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report OCRBench_v2 en 66.8, zh 63.5, no overall (Table 2) Capabilities
Nov 6, 2025 Nemotron Nano 12B v2 VL technical report 54.8 OCRBenchV2 EN, reasoning on, ZH 39.8 (Table 3). Reasoning off EN 62.0, ZH 44.2 Capabilities
Apr 30, 2025 Amazon Nova Premier technical report 56.9 OCRBench-v2 (Table 1) Capabilities
Mar 26, 2025 Qwen2.5-Omni-7B technical report 57.8 OCRBench_V2 en (Table 5) Capabilities
Feb 19, 2025 Qwen2.5-VL-72B technical report OCRBench_v2 en 61.5, zh 63.7, no overall (Table 5) Capabilities

The lab's number comes out ahead in 5 of 5 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
LFM2.5-VL-3B 47.5% OCRBench v2 (En) 42.6% Leaderboard +4.9 pts
Llama-3.1-Nemotron-Nano-VL-8B-V1 60.1% OCRBenchV2 English 60.1 / Chinese 37.9 56.4% Leaderboard +3.7 pts
Nemotron 3 Nano Omni 30B-A3B 67% OCRBenchV2 (EN) 65.8% Leaderboard +1.2 pts
Nemotron Nano 12B v2 VL 62% OCRBenchv2 (English) 61.2% Leaderboard +0.8 pts
Nemotron 3.5 Super VL 68.1% OCRBench v2, EN 68.12 / ZH 59.15 67.4% Leaderboard +0.7 pts

2 logged changes that move scores

Changes by the maintainers, so a score from before a change may not compare with one from after it. Newest first, each with the maintainers' own source. The list may be incomplete. What we log.

  1. Grader changed July 5th, 2025

    The maintainers corrected English answer-length routing. Longer answers now receive edit-distance scoring, while short answers use substring matching. Source

  2. Tasks changed June 21st, 2025

    The maintainers introduced a private evaluation dataset and quarterly leaderboard updates. Private-set results measure a different question set from the public release. Source

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 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (26 entries), launch posts (20 scores) and system cards (14 scores).
  • Pass Models scored 34 models we track, plus 11 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 Qwen 3.8 Max (#30).
  • Pass Headroom The best is 74.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on 99franklin.github.io: 15 rows for models we track and 11 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims All 5 pairs agree within 10 points.
  • Pass Known issues No published audit found problems.

How we check