OpenVLM ChartQA Test: leaderboard
ChartQA test split from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): question answering over charts, scored by relaxed accuracy (numeric answers within a small tolerance) in percent, higher is better. Exact-zero rows are failed runs and are dropped.
Metric: Relaxed Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 108 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 VL 72B | 88.2 | #450 |
| 2 | InternVL3-14B | 87.8 | #506 |
| 3 | InternVL3-8B | 85.8 | #624 |
| 4 | MiniCPM-V-2.6 | 82.4 | #827 |
| 5 | Pixtral-12B | 72.2 | #805 |
| 6 | Gemma 3 12B | 37.2 | #674 |
| 7 | GPT-4.1 Mini | 33.7 | #363 |
| 8 | Gemma 3 4B | 33.7 | #1088 |
| 9 | GPT-4.1 (2025-04-14) | 33.3 | #257 |
| 10 | Aquila-VL-2B | 32 | #1216 |
| 11 | GPT-4.1 Nano | 30.3 | #730 |
| 12 | Gemma 3 27B | 30.1 | #606 |
| 13 | Gemini 1.5 Flash | 29.7 | #620 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-chartqa-test · How It Works · Data refreshed daily, snapshot 2026-11-05.