LLM Stats (CharXiv-R): leaderboard
CharXiv pairs 2,323 charts from arXiv papers with human curated questions. Its reasoning set, CharXiv-R, asks one question per chart that combines several chart elements, with a text or number answer, and a judge model marks each answer right or wrong instead of exact match. This board collects accuracy reported by model providers and unverified by llm-stats, higher is better. Source: arxiv.org/abs/2406.18521.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 60 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.1 | 88.4 | #52 |
| 2 | Muse Spark | 86.4 | #64 |
| 3 | Gemini 3.8 Flash | 86.2 | #21 |
| 4 | Gemini 3.5 Flash | 84.2 | #60 |
| 5 | MiniMax-M3.1-Flash | 82.5 | #33 |
| 6 | GPT-5.2 | 82.1 | #110 |
| 7 | GPT-5.5 Instant | 81.6 | #129 |
| 8 | Qwen 3.6 Plus | 81.5 | #120 |
| 9 | Gemini 3 Pro | 81.4 | #80 |
| 10 | GPT-5 | 81.1 | #98 |
| 11 | MiMo-V2.5 | 81 | #196 |
| 12 | Nemotron 3.5 Super VL (BF16) | 80.82 | #165 |
| 13 | Gemini 3 Flash | 80.3 | #99 |
| 14 | Qwen 3.5 27B | 79.5 | #187 |
| 15 | Muse Glimmer 30B | 78.8 | #178 |
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-r · How It Works · Data refreshed daily, snapshot 2026-10-20.