Benchmarks
Language performance
How well does each AI model perform when prompted in different languages? Within each language, models are ranked by head-to-head comparison on the same native prompts — 1000 is the average of the models measured in that language.
Tests run weekly · Models with image/audio/TTS capabilities excluded
English
prompts: 2 · 397 scored runs · English
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 95%±5 | 1 |
| 1 | gpt-oss-120bOVH AI Endpoints (GRA) | 90%±5 | 8 |
| 1 | Claude Opus 4.8Anthropic | 89%±5 | 6 |
| 1 | gpt-5.4-nano-2026-03-17OpenAI | 87%±6 | 4 |
| 1 | Claude Opus 4.7Anthropic | 84%±5 | 9 |
Nederlands
prompts: 2 · 330 scored runs · Dutch
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Claude Fable 5Anthropic | 85%±6 | 3 |
| 1 | Claude Sonnet 5Anthropic | 85%±6 | 2 |
| 1 | gpt-5.2OpenAI | 85%±6 | 4 |
| 1 | gpt-5.2-2025-12-11OpenAI | 85%±6 | 4 |
| 1 | gpt-5.5-2026-04-23OpenAI | 85%±6 | 4 |
Deutsch
prompts: 1 · 327 scored runs · German
Only 1 prompt in this language so far — this shows the evidence collected, not a ranking. More prompts are needed before the order means anything.
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 66%±12 | 2 |
| 1 | Claude Opus 4.5Anthropic | 60%±9 | 5 |
| 1 | Claude Opus 4.6Anthropic | 60%±9 | 5 |
| 1 | Claude Opus 4.7Anthropic | 60%±9 | 5 |
| 1 | Claude Opus 4.8Anthropic | 60%±9 | 3 |
Français
prompts: 1 · 276 scored runs · French
Only 1 prompt in this language so far — this shows the evidence collected, not a ranking. More prompts are needed before the order means anything.
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Gemini Flash-Lite LatestGoogle Gemini | 82%±9 | 6 |
| 1 | gpt-5.1OpenAI | 82%±9 | 3 |
| 1 | o3OpenAI | 82%±9 | 3 |
| 1 | Gemini 3.1 Flash LiteGoogle Gemini | 82%±9 | 6 |
| 1 | gpt-5.4-2026-03-05OpenAI | 82%±9 | 3 |
Español
prompts: 1 · 270 scored runs · Spanish
Only 1 prompt in this language so far — this shows the evidence collected, not a ranking. More prompts are needed before the order means anything.
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Gemini Flash-Lite LatestGoogle Gemini | 82%±7 | 5 |
| 1 | gpt-4.1-miniOpenAI | 82%±7 | 6 |
| 1 | gpt-4.1-mini-2025-04-14OpenAI | 82%±7 | 5 |
| 1 | gpt-4o-miniOpenAI | 82%±7 | 5 |
| 1 | gpt-4o-mini-2024-07-18OpenAI | 82%±7 | 6 |
Türkçe
prompts: 1 · 89 scored runs · Turkish
Only 1 prompt in this language so far — this shows the evidence collected, not a ranking. More prompts are needed before the order means anything.
| # | Model | Win rate | Runs |
|---|---|---|---|
| 1 | Claude Fable 5Anthropic | 57%±12 | 2 |
| 1 | Claude Haiku 4.5Anthropic | 57%±12 | 2 |
| 1 | Claude Opus 4.5Anthropic | 57%±12 | 2 |
| 1 | Claude Opus 4.6Anthropic | 57%±12 | 2 |
| 1 | Claude Opus 4.7Anthropic | 57%±12 | 2 |