DiscoX: leaderboard
DiscoX — discourse-level cross-lingual translation and coherence evaluation; overall score across discourse phenomena.
Metric: Overall Score. Source: randomtutu.github.io. Saturation forecast: Around January 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 76.66 | #98 (GPT-5) |
| 2 | Gemini 2.5 Pro | 71.25 | #152 |
| 3 | Qwen 3 235B A22B | 59.66 | #313 |
| 4 | Kimi K2 | 55.8 | #243 |
| 5 | O3 (High) | 55.57 | #124 (O3) |
| 6 | O4 Mini (High) | 55.09 | #180 (O4 Mini) |
| 7 | GPT-4.1 | 49.65 | #248 |
| 8 | DeepSeek V3 | 49.6 | #321 |
| 9 | DeepSeek R1 | 46.06 | #251 |
| 10 | Gemini 2.5 Flash Lite | 44.01 | #422 |
| 11 | Grok 4 | 43.82 | #181 |
| 12 | GPT-4o | 39.93 | #343 |
| 13 | Qwen 3 14B | 39.36 | #537 |
| 14 | Qwen 3 8B | 28.37 | #674 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=discox · How It Works · Data refreshed daily, snapshot 2026-10-20.