CulturALL: leaderboard
Multilingual and multicultural competence on grounded everyday tasks from CulturALL (2026): 2,610 scenario questions in 14 languages from 51 regions across 16 topics such as travel, shopping and local customs, written with native-speaker annotators and made deliberately hard. Without web search; accuracy in percent judged against reference answers, higher is better.
Metric: Accuracy (%) without web search on all 2,610 CulturALL items (grounded everyday tasks in 14 languages and 51 regions written with expert annotators; items that every one of the paper's 15 model settings solved were removed), zero-shot, answers judged correct or incorrect by gpt-4o-2024-11-20 against the reference; higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 7 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 2.5 Pro | 37.89 | #276 | |
| 2 | GPT-5 (2025-08-07) | High | 37.59 | |
| 3 | GPT-5 | 37.2 | #178 | |
| 4 | Claude Opus 4 (20250514) | Thinking | 36.7 | #297 |
| 5 | Claude Sonnet 4 (20250514) | Thinking | 32.76 | #405 |
| 6 | Qwen 3 235B A22B | 23.68 | #628 | |
| 7 | CulturALL qwen-max (checkpoint unspecified) | 18.97 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=culturall · How It Works · Data refreshed daily, snapshot 2026-11-21.