LiT (Lost in Translation) Robustness: 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 built around classical round-trip pitfalls, averaged over five categories, whose round trip scores at least 80, 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); mean of the five category rates on the robustness subset of 60 sentences built around classical round-trip translation pitfalls; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 FlashNon-reasoning71#175
2Qwen 3.5 397B A17B63.5#264
3GLM-559.8#269
4Gemma 4 31B (IT)Thinking55.8#350
5Kimi K2 (Thinking)54.6
6GLM-4.748.8#380
7Qwen 3 235B A22B 2507 (Thinking)47.5#466
8Qwen 3.5 35B A3B43.5#492
9DeepSeek V3.2 ExpThinking41#439
10MiniMax-M2.538.8#600
11Gemma 3 27B (IT)37.5#1324
12GPT-OSS-120BHigh34.6#704
13Kimi K232.1#443
14Qwen 3 235B A22B 2507 Instruct31.9#616
15MiMo-V2-FlashNon-reasoning24.4#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 · How It Works · Data refreshed daily, snapshot 2026-11-21.