EsoBench — leaderboard

Can models learn an unknown programming language from scratch? 30 tasks across 6 difficulty levels with only 1 example and 50 experimental attempts. No model has fully solved it - top score is 29.4/100.

Metric: Score. Source: caseys-evals.com. Status: saturation imminent. 38 models tracked.

Top models

#ModelScore
1Qwen 3 235B A22B (Thinking)29.4
2GPT-528.13
3GPT-5.2 (High)26.87
4O4 Mini (2025-04-16) (High)26.8
5GPT-5 Mini23.73
6O4 Mini (2025-04-16)23.07
7Grok 421.4
8Claude Opus 4.5 (20251101)20.93
9GPT-5.1 (High)19.33
10Claude Opus 4.1 (20250805)19.27
11Gemini 3 Pro (Preview)17.73
12Claude Sonnet 4.617.4
13Grok 316.8
14Claude Opus 4 (20250514)16.8
15O3 Mini (2025-01-31)16.67

Interactive version: theaggregate.ai/benchmark?slug=esobench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.