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. Status: years away from saturation. 37 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 55.9 |
| 2 | Gemini 2.5 Pro | 50.65 |
| 3 | GPT-4o | 43.4 |
| 4 | DeepSeek V3 | 35.94 |
| 5 | Gemma 3 27B | 35.65 |
| 6 | Gemini 1.5 Pro | 33.49 |
| 7 | Qwen 2.5 72B | 33.2 |
| 8 | Llama 3.3 70B | 32.74 |
| 9 | Gemma 2 27B | 32.36 |
| 10 | Llama 3.1 405B | 32.31 |
| 11 | Qwen 2.5 32B | 31.65 |
| 12 | Claude Opus 4 | 31.29 |
| 13 | Llama 3.1 70B | 31.27 |
| 14 | Qwen 3 14B | 30.79 |
| 15 | Qwen 2 72B | 30.73 |
Interactive version: theaggregate.ai/benchmark?slug=asciieval · How the rankings work · Data refreshed daily, snapshot 2026-07-22.