ChartDiff: leaderboard
Metric: GPT Score (0 to 5) on all test pairs, ChartDiff's 1,500 test chart pairs (two charts rendered side by side as one image), zero-shot; a GPT-5.4 judge compares the model's comparison summary with the underlying CSV data and the human-verified reference summary and gives an integer from 0 (fail) to 5 (excellent) for accuracy, completeness, faithfulness and clarity; mean score; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 4.95 | #76 |
| 2 | Gemini 3.1 Pro (Preview) | 4.86 | #54 |
| 3 | GPT-5.4 Mini | 4.82 | #202 |
| 4 | Gemini 3.1 Flash Lite | 4.63 | #174 |
| 5 | Claude Sonnet 4.6 | 4.58 | #85 |
| 6 | Qwen 3.5 397B A17B | 4.54 | #141 |
| 7 | GPT-4o | 4.23 | #333 |
| 8 | Qwen 3.5 9B | 3.65 | #363 |
| 9 | Qwen 2.5 VL 7B Instruct | 3.18 | #643 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=chartdiff · How It Works · Data refreshed daily, snapshot 2026-10-11.