Macaron - Multiple Choice (Local Language): leaderboard
Metric: Accuracy (%) on the local-language multiple-choice version (20 languages and dialects, including Amharic, Yoruba, Zulu, Kyrgyz and four Arabic dialects) of the Macaron scenarios: 1,977 culturally grounded questions written by native annotators from 100 language-agnostic templates (7 reasoning types, 22 cultural aspects) for 20 cultural contexts and languages; zero-shot; open-weight models up to 8B scored by option log-probability, API models by a final-line JSON answer; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 88.3 | #145 |
| 2 | Gemini 3 Flash (Preview) (Non-reasoning) | 87 | #78 (Gemini 3 Flash (Preview)) |
| 3 | Claude Opus 4.5 | 80.2 | #79 |
| 4 | Gemini 2.5 Flash (Non-reasoning) | 80.2 | #237 (Gemini 2.5 Flash) |
| 5 | GPT-5 Chat | 78.2 | #233 |
| 6 | DeepSeek V3.1 (Thinking) | 75.6 | #260 (DeepSeek V3.1) |
| 7 | DeepSeek V3.1 (Non-reasoning) | 68.4 | #260 (DeepSeek V3.1) |
| 8 | Qwen 3 235B A22B 2507 Instruct | 68.3 | #291 |
| 9 | Llama 4 Maverick | 67.5 | #451 |
| 10 | Gemini 2.5 Flash (Thinking) | 65.9 | #237 (Gemini 2.5 Flash) |
| 11 | GPT-4o Mini | 65.3 | #588 |
| 12 | Claude Haiku 4.5 | 63.8 | #271 |
| 13 | Llama 3.3 70B Instruct | 62.6 | #520 |
| 14 | aya-expanse-8B | 48.7 | #921 |
| 15 | Qwen 2.5 7B Instruct | 46.9 | #846 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=macaron-multiple-choice-local-language · How It Works · Data refreshed daily, snapshot 2026-10-11.