LLM Stats (CharXiv-D): leaderboard
Descriptive questions about 2,323 complex scientific charts from arXiv, requiring identification of specific data points, labels, and visual elements.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.1 Pro | 95.5 | #67 |
| 2 | Seed 2.1 Turbo | 94.6 | #86 |
| 3 | Seed 2.0 Mini | 91.9 | #276 |
| 4 | Qwen 3 VL 32B Instruct | 90.5 | #283 |
| 5 | Qwen 3 VL 32B (Thinking) | 90.2 | #294 (Qwen 3 VL 32B) |
| 6 | GPT-4.5 | 90 | #274 |
| 7 | GPT-4.1 Mini | 88.4 | #355 |
| 8 | Command A+ | 88 | #413 |
| 9 | GPT-4.1 | 87.9 | #248 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 86.9 | #351 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 3 VL 8B (Thinking) | 85.9 | |
| 12 | Qwen 3 VL 30B A3B Instruct | 85.5 | #384 |
| 13 | GPT-4o | 85.3 | #343 |
| 14 | Qwen 3 VL 4B (Thinking) | 83.9 | #480 (Qwen 3 VL 4B) |
| 15 | Qwen 3 VL 8B Instruct | 83 | #420 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-charxiv-d · How It Works · Data refreshed daily, snapshot 2026-10-20.