Edit2TikZ - Edit Correctness Score: leaderboard
Edit2TikZ (arXiv 2608.13441v1, Table 3): 1,548 scientific-figure edits where a multimodal model gets a rendered figure and an instruction and must write compilable TikZ for the revised figure. Edit Correctness Score averages the judged completion of each labelled edit unit, 0-100, zero for render failures; GPT-5.6-Terra is the judge and is also evaluated. Higher is better.
Metric: ECS (0-100; judged per edit unit, zero for render failures). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 75.02 | #64 |
| 2 | Claude Opus 4.8 | 64.98 | #39 |
| 3 | GPT-5.6 Sol | 64.1 | #22 |
| 4 | GPT-5.6 Terra | 61.04 | #40 |
| 5 | Claude Sonnet 4.6 | 59.55 | #94 |
| 6 | GPT-5.6 Luna | 54.79 | #79 |
| 7 | Seed 2.1 Pro | 49.88 | #73 |
| 8 | Qwen 3.7 Plus | 49.04 | #98 |
| 9 | Qwen 3.6 27B | 48.5 | #182 |
| 10 | gemma-4-E4B-it | 30.18 | #577 |
| 11 | Qwen 3.5 9B | 24.99 | #391 |
| 12 | Qwen 3.5 4B | 22.54 | #514 |
| 13 | Gemma 4 E2B (IT) | 2.97 | #779 |
| 14 | Qwen 3.5 2B | 2.78 | #1114 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=edit2tikz-edit-correctness-score · How It Works · Data refreshed daily, snapshot 2026-11-05.