ASCIIEval: leaderboard
Benchmarks LLMs' ability to visually perceive and interpret ASCII art. Tests 37 text-only models on macro-average accuracy across visual concepts rendered as ASCII characters.
Metric: Macro-Avg Accuracy (%). Source: github.com. Saturation forecast: Around 2031. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 55.9 | #98 |
| 2 | Gemini 2.5 Pro | 50.65 | #152 |
| 3 | GPT-4o | 43.4 | #343 |
| 4 | DeepSeek V3 | 35.94 | #321 |
| 5 | Gemma 3 27B | 35.65 | #607 |
| 6 | Gemini 1.5 Pro | 33.49 | #445 |
| 7 | Qwen 2.5 72B | 33.2 | #499 |
| 8 | Llama 3.3 70B | 32.74 | #579 |
| 9 | Gemma 2 27B | 32.36 | #806 |
| 10 | Llama 3.1 405B | 32.31 | #466 |
| 11 | Qwen 2.5 32B | 31.65 | #546 |
| 12 | Claude Opus 4 | 31.29 | #163 |
| 13 | Llama 3.1 70B | 31.27 | #592 |
| 14 | Qwen 3 14B | 30.79 | #537 |
| 15 | Qwen 2 72B | 30.73 | #521 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=asciieval · How It Works · Data refreshed daily, snapshot 2026-10-20.