Benchmark Almanac

OCR Arena

Claude Opus 4.5 1,791 leads since Nov 2025 Not in model cards

Users upload a document, compare two anonymous OCR or vision-language models side by side, and vote. Votes become Elo ratings.

Launched with more than 10 models on Hacker News on 2025-11-21 (216 points).

Made by Extend Released Nov 2025 Documents, charts and text in images All tasks public Last model added Feb 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Ranked by votes People pick the better of two answers, as in an arena, and the votes are turned into ratings.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
Leaderboard ocrarena.ai
18
Entries on its own leaderboard, 11 models we track

Claude Opus 4.5 leads with 1,791 among models released by Feb 2026

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

The newest model with a score here came out in Feb 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

18 entries. Scroll sideways for the rest. Bars start at 1,300.

  1. 1,791 Claude Opus 4.5
  2. 1,770 Gemini 3 Flash
  3. 1,713 Claude Opus 4.6
  4. 1,701 Gemini 2.5 Pro
  5. 1,700 Gemini 2.5 Flash
  6. 1,631 Claude Sonnet 4.5
  7. 1,536 Qwen3-VL-8B
  8. 1,517 Mistral OCR v3
  9. 1,515 dots.ocr
  10. 1,464 Nanonets2-3B
  11. 1,421 GLM-OCR
  12. 1,397 olmOCR 2
  13. 1,382 GPT-5
  14. 1,377 DeepSeek OCR
  15. 1,367 GPT-5.1
  16. 1,367 Gemini 3 Pro
  17. 1,359 GPT-5.2
  18. 1,327 Iris

Bars start at 1,300.

  1. 1 Claude Opus 4.5 low 1,791
  2. 2 Gemini 3 Flash 1,770
  3. 3 Claude Opus 4.6 adaptive 1,713
  4. 4 Gemini 2.5 Pro Jun 2025 1,701
  5. 5 Gemini 2.5 Flash Sep 2025 1,700
  6. 6 Claude Sonnet 4.5 1,631
  7. 7 Qwen3-VL-8B 1,536
  8. 8 Mistral OCR v3 1,517
  9. 9 dots.ocr 1,515
  10. 10 Nanonets2-3B 1,464
  11. 11 GLM-OCR 1,421
  12. 12 olmOCR 2 1,397
  13. 13 GPT-5 low 1,382
  14. 14 DeepSeek OCR 1,377
  15. 15 GPT-5.1 medium 1,367
  16. 16 Gemini 3 Pro 1,367
  17. 17 GPT-5.2 none 1,359
  18. 18 Iris 1,327
#EntryScoreSettingDateSource
1 Claude Opus 4.5 1,791 low Nov 2025 Official leaderboard
2 Gemini 3 Flash 1,770 Dec 2025 Official leaderboard
3 Claude Opus 4.6 1,713 adaptive Feb 2026 Official leaderboard
4 Gemini 2.5 Pro Jun 2025 1,701 Jun 2025 Official leaderboard
5 Gemini 2.5 Flash Sep 2025 1,700 Sep 2025 Official leaderboard
6 Claude Sonnet 4.5 1,631 Sep 2025 Official leaderboard
7 Qwen3-VL-8B 1,536 Official leaderboard
8 Mistral OCR v3 1,517 Official leaderboard
9 dots.ocr 1,515 Official leaderboard
10 Nanonets2-3B 1,464 Official leaderboard
11 GLM-OCR 1,421 Jan 2026 Official leaderboard
12 olmOCR 2 1,397 Official leaderboard
13 GPT-5 1,382 low Aug 2025 Official leaderboard
14 DeepSeek OCR 1,377 Official leaderboard
15 GPT-5.1 1,367 medium Nov 2025 Official leaderboard
16 Gemini 3 Pro 1,367 Nov 2025 Official leaderboard
17 GPT-5.2 1,359 none Dec 2025 Official leaderboard
18 Iris 1,327 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Opus 4.5
low medium
1,791
Claude Opus 4.6
adaptive low medium
1,713
GPT-5
low
1,382
GPT-5.1
low
1,367
GPT-5.2
none
1,359
ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.5 low 1,791
Gemini 3 Flash 1,770
Claude Opus 4.6 adaptive 1,713
Claude Opus 4.5 medium 1,709
Claude Opus 4.6 medium 1,707
Gemini 2.5 Pro Jun 2025 1,701
Gemini 2.5 Flash Sep 2025 1,700
Claude Opus 4.6 low 1,667
Claude Sonnet 4.5 1,631
GLM-OCR 1,421
GPT-5 low 1,382
GPT-5.1 medium 1,367
Gemini 3 Pro 1,367
GPT-5.1 low 1,362
GPT-5 medium 1,360
GPT-5.2 none 1,359
GPT-5.2 medium 1,324

2 models have held number one since its release in Nov 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind Anthropic
Own leaderboard Not tested since Feb 2026
Claude Opus 4.5 1,791

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
1,300 1,400 1,500 1,600 1,700 1,800 2025 2026 1,300 1,400 1,500 1,600 1,700 1,800 2025 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Vision

All 234 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. Kimi K3.1 Pro 84.1%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%
  • DocVQA In use Models answer 50,000 questions about 12,000 scanned document images by reading their text and layout. Scoring is ANLS, a string similarity to reference answers. Seed1.5-VL 96.9%