multimodal benchmark
TextVQA
TextVQA contains 45,336 questions on 28,408 images that require reasoning about text to answer. Introduced to benchmark VQA models' ability to read and reason about text within images, particularly for assistive technologies for visually impaired users. The dataset addresses the gap where existing VQA datasets had few text-based questions or were too small.
Updated Aug 7, 2026
Published models15
Registry coverage15
MetricScore
EvidenceB
TextVQA leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 85.5% | 100.0% | 15 | C | |
| 2 | AC | 84.9% | 92.9% | 15 | C | |
| 3 | AC | 84.4% | 85.7% | 15 | C | |
| 4 | DE | 84.2% | 78.6% | 15 | C | |
| 5 | DE | 83.4% | 71.4% | 15 | C | |
| 6 | AM Nova ProAmazon | 81.5% | 64.3% | 15 | C | |
| 7 | DE | 80.7% | 57.1% | 15 | C | |
| 8 | AMNova LiteAmazon | 80.2% | 50.0% | 15 | C | |
| 9 | XA | 78.1% | 42.9% | 15 | C | |
| 10 | MI | 75.6% | 35.7% | 15 | C | |
| 11 | ME | 73.5% | 28.6% | 15 | C | |
| 12 | MI | 72.0% | 21.4% | 15 | C | |
| 13 | GO | 67.7% | 14.3% | 15 | C | |
| 14 | GO | 65.1% | 7.1% | 15 | C | |
| 15 | GO | 57.8% | 0.0% | 15 | C |
Score distribution
Top published rows on the benchmark's original scale.
TextVQA highlights
The top published results on this benchmark's own scale.
What is TextVQA?
Definition and scoring fields from the benchmark registry.
TextVQA contains 45,336 questions on 28,408 images that require reasoning about text to answer. Introduced to benchmark VQA models' ability to read and reason about text within images, particularly for assistive technologies for visually impaired users. The dataset addresses the gap where existing VQA datasets had few text-based questions or were too small.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- TextVQA
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- textvqa|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about TextVQA.
Which model scores highest on TextVQA?
Qwen2-VL-72B-Instruct is currently ranked first with 85.5%.
What does TextVQA measure?
TextVQA contains 45,336 questions on 28,408 images that require reasoning about text to answer. Introduced to benchmark VQA models' ability to read and reason about text within images, particularly for assistive technologies for visually impaired users. The dataset addresses the gap where existing VQA datasets had few text-based questions or were too small.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
15 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.