LiT (Lost in Translation) Robustness - Register and Tone Shifts: leaderboard
LiT, Lost in Translation (2026), measures multilingual proficiency by round-trip translation: English passages pass through a sequence of four languages and back, and an LLM judge scores the result against the source with MQM error weights. Share of robustness sentences with shifts of register and tone whose round trips score at least 80, averaged over eight translation sequences, in percent; higher is better.
Metric: MQM>=80 rate (%): share of round-trip translations whose back-translation to English scores at least 80 on an MQM error scale (minor -1, major -5, critical -25) from a Grok 4.1 Fast judge comparing it with the English source; each passage is translated serially through a four-language sequence and back, averaged over eight sequences (high, medium and low-resource languages); robustness subset: register and tone shifts; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Flash | Non-reasoning | 60.4 | #175 |
| 2 | Qwen 3.5 397B A17B | 58.3 | #264 | |
| 3 | GLM-5 | 57.3 | #269 | |
| 4 | Gemma 4 31B (IT) | Thinking | 55.2 | #350 |
| 5 | Kimi K2 (Thinking) | 50 | ||
| 6 | Qwen 3.5 35B A3B | 43.8 | #492 | |
| 7 | GLM-4.7 | 43.8 | #380 | |
| 8 | GPT-OSS-120B | High | 38.5 | #704 |
| 9 | Gemma 3 27B (IT) | 37.5 | #1324 | |
| 10 | Qwen 3 235B A22B 2507 (Thinking) | 37.5 | #466 | |
| 11 | DeepSeek V3.2 Exp | Thinking | 34.4 | #439 |
| 12 | Qwen 3 235B A22B 2507 Instruct | 32.3 | #616 | |
| 13 | MiniMax-M2.5 | 32.3 | #600 | |
| 14 | Kimi K2 | 24.6 | #443 | |
| 15 | MiMo-V2-Flash | Non-reasoning | 13.5 | #684 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=lit-lost-in-translation-robustness-register-and-tone-shifts · How It Works · Data refreshed daily, snapshot 2026-11-21.