Evals for Every Language - MGSM: leaderboard
LanguageBench MGSM accuracy averaged across available languages and scaled to 0-100. Translation uses spBLEU only, without mixing chrF; higher is better.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 70 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 96.62 | #37 |
| 2 | Claude Opus 4.7 | 95.57 | #49 |
| 3 | Gemini 3.1 Pro (Preview) | 93.12 | #58 |
| 4 | Claude Sonnet 4.6 | 91.89 | #90 |
| 5 | GPT-5.5 | 90.21 | #31 |
| 6 | Grok 4.20 | 87.39 | #175 |
| 7 | GPT-5.4 | 86.98 | #79 |
| 8 | Claude Sonnet 4.5 | 86.76 | #142 |
| 9 | Qwen 3.6 Plus | 86.72 | #120 |
| 10 | Gemini 3.1 Flash Lite | 85.23 | #191 |
| 11 | DeepSeek V3.1 | 85.22 | #264 |
| 12 | GPT-5.3 | 84.79 | #103 |
| 13 | GPT-5.2 | 84.65 | #110 |
| 14 | Gemini 2.5 Pro | 83.92 | #152 |
| 15 | GPT-5 | 82.71 | #98 |
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-mgsm · How It Works · Data refreshed daily, snapshot 2026-10-20.