Evals for Every Language: leaderboard
LanguageBench overall: equal-weight mean of six task metrics after averaging each over available languages. Uses translation-from and translation-to spBLEU, classification, MMLU, ARC and MGSM accuracy; all six are required.
Metric: Six-task Macro Average (%). Source: huggingface.co. Saturation forecast: Around January 2027. 69 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 75.33 | #58 |
| 2 | Claude Opus 4.7 | 73.37 | #49 |
| 3 | Claude Opus 4.8 | 72.91 | #37 |
| 4 | Gemini 3.1 Flash Lite | 71.35 | #191 |
| 5 | Claude Sonnet 4.6 | 70.91 | #90 |
| 6 | GPT-5.5 | 70.76 | #31 |
| 7 | Claude Sonnet 4.5 | 70.37 | #142 |
| 8 | GPT-5.3 | 69.77 | #103 |
| 9 | Grok 4.20 | 69.62 | #175 |
| 10 | GPT-5 | 69.24 | #98 |
| 11 | Gemini 2.5 Pro | 69.05 | #152 |
| 12 | Claude Sonnet 4 | 68.65 | #199 |
| 13 | GPT-5.2 | 68.47 | #110 |
| 14 | GPT-5.4 | 68.4 | #79 |
| 15 | Qwen 3.6 Plus | 68.26 | #120 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=evals-for-every-language · How It Works · Data refreshed daily, snapshot 2026-10-20.