Benchmark Almanac

OmniAI OCR Benchmark

STEP3-VL-10B 77%
Llama 4 Maverick 82.3% leads since April 2025
Llama 4 Maverick 82.3% leads since April 2025
Only in system cards

Multimodal models and OCR providers extract text and JSON fields from documents. Accuracy is measured on both plain text and structured extraction.

About 75 likes on Hugging Face.

Made by
OmniAI
Released
February 2025
Kind
Documents, charts and text in images
Level
Everyday
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
Llama 4 Maverick #153 on the capability index
Last in a system card
January 2026
Last model added
January 2026
Scores refreshed
November 12th, 2026
1
Model scored in lab cards
1
System card reports it, from 1 lab

STEP3-VL-10B leads with 77%

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

Llama 4 Maverick leads with 82.3% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.

Llama 4 Maverick leads with 82.3%, 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.

  1. 77.0 STEP3-VL-10B
  1. 1 STEP3-VL-10B 77%

Source: Labs’ model cards

Source: Official OmniAI OCR Benchmark leaderboard

Sources: Official OmniAI OCR Benchmark leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 STEP3-VL-10B 77% Jan 2026 Lab model card
#ModelScoreSettingDateSource
1 Llama 4 Maverick 82.3% Apr 2025 Official leaderboard
2 Qwen2.5-VL-72B 75.2% Jan 2025 Official leaderboard
3 Llama 3.2 90B 54.2% Sep 2024 Official leaderboard
4 Gemma 3 27B 42.9% Mar 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Llama 4 Maverick 82.3% Apr 2025 Official leaderboard
2 STEP3-VL-10B 77% Jan 2026 Lab model card
3 Qwen2.5-VL-72B 75.2% Jan 2025 Official leaderboard
4 Llama 3.2 90B 54.2% Sep 2024 Official leaderboard
5 Gemma 3 27B 42.9% Mar 2025 Official leaderboard

The benchmark's own leaderboard has 14 entries, 4 of them models we track. Switch the scores above to rank them together.

JSON accuracy versus page cost

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

60% 80% 100% $0$5$10$15$20 Cost per 1,000 pages ($), lower is better JSON accuracy Llama 4 Maverick: JSON accuracy 82.3%, cost per 1,000 pages $1.98 Azure (OCR service): JSON accuracy 85.1%, cost per 1,000 pages $10. Not in our model list GPT-4o (snapshot unspecified): JSON accuracy 75.5%, cost per 1,000 pages $18.4. Not in our model list AWS Textract (OCR service): JSON accuracy 74.3%, cost per 1,000 pages $4. Not in our model list Claude Sonnet 3.5 (snapshot unspecified): JSON accuracy 69.3%, cost per 1,000 pages $19.9. Not in our model list Google Document AI (OCR service): JSON accuracy 67.8%, cost per 1,000 pages $1.5. Not in our model list Unstructured (OCR service): JSON accuracy 50.8%, cost per 1,000 pages $20. Not in our model list OmniAI (OCR service): JSON accuracy 91.7%, cost per 1,000 pages $10. Not in our model list Gemini 2.0 Flash (snapshot unspecified): JSON accuracy 86.1%, cost per 1,000 pages $1.12. Not in our model list Mistral OCR (snapshot unspecified): JSON accuracy 72.2%, cost per 1,000 pages $1. Not in our model list GPT-4o Mini (snapshot unspecified): JSON accuracy 64.8%, cost per 1,000 pages $0.97. Not in our model list OmniAI (OCR service)Gemini 2.0 Flash (snapshot unspecified)Mistral OCR (snapshot unspecified)GPT-4o Mini (snapshot unspecified) 60% 80% 100% $0$5$10$15$20 Cost per 1,000 pages ($), lower is better JSON accuracy Llama 4 Maverick: JSON accuracy 82.3%, cost per 1,000 pages $1.98 Azure (OCR service): JSON accuracy 85.1%, cost per 1,000 pages $10. Not in our model list GPT-4o (snapshot unspecified): JSON accuracy 75.5%, cost per 1,000 pages $18.4. Not in our model list AWS Textract (OCR service): JSON accuracy 74.3%, cost per 1,000 pages $4. Not in our model list Claude Sonnet 3.5 (snapshot unspecified): JSON accuracy 69.3%, cost per 1,000 pages $19.9. Not in our model list Google Document AI (OCR service): JSON accuracy 67.8%, cost per 1,000 pages $1.5. Not in our model list Unstructured (OCR service): JSON accuracy 50.8%, cost per 1,000 pages $20. Not in our model list OmniAI (OCR service): JSON accuracy 91.7%, cost per 1,000 pages $10. Not in our model list Gemini 2.0 Flash (snapshot unspecified): JSON accuracy 86.1%, cost per 1,000 pages $1.12. Not in our model list Mistral OCR (snapshot unspecified): JSON accuracy 72.2%, cost per 1,000 pages $1. Not in our model list GPT-4o Mini (snapshot unspecified): JSON accuracy 64.8%, cost per 1,000 pages $0.97. Not in our model list OmniAI (OCR service)Gemini 2.0 Flash (snapshot unspecified)Mistral OCR (snapshot unspecified)GPT-4o Mini (snapshot unspecified)
ModelSettingJSON accuracyCost per 1,000 pages
OmniAI (OCR service) 91.7% $10
Gemini 2.0 Flash (snapshot unspecified) 86.1% $1.12
Azure (OCR service) 85.1% $10
Llama 4 Maverick 82.3% $1.98
GPT-4o (snapshot unspecified) 75.5% $18.37
AWS Textract (OCR service) 74.3% $4
Mistral OCR (snapshot unspecified) 72.2% $1
Claude Sonnet 3.5 (snapshot unspecified) 69.3% $19.93
Google Document AI (OCR service) 67.8% $1.5
GPT-4o Mini (snapshot unspecified) 64.8% $0.97
Unstructured (OCR service) 50.8% $20

Number one over time 2 models have held number one across all independent runs since its release in February 2025 2 models have held number one with claims included since its release in February 2025

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

Qwen Meta
Own leaderboard Not tested since Apr 2025
Llama 4 Maverick 82.3%

Source: Official OmniAI OCR Benchmark leaderboard

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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Llama 4 Maverick Qwen2.5-VL-72B Llama 3.2 90B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Llama 4 Maverick Qwen2.5-VL-72B

Sources: Official OmniAI OCR Benchmark leaderboard and labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

One system card reports it

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
Jan 14, 2026 STEP3-VL-10B technical report 77 OmniOCR (OmniAI), SeRe (Tables 1 and 3). PaCoRe 78.14 Capabilities

More in Vision

All 281 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. 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%
  • 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%
  • 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%