TransClean Generation - Prompt 1 - Wrong Language: leaderboard
The TransClean paper measures unwanted extra text and wrong-language outputs when models translate a multilingual source collection. These are translation-generation diagnostics preceding TransClean's separate extraction task. Prompt 0 is a language-labelled completion, while prompts 1 and 2 explicitly request translation; error rates retain their lower-is-better meaning despite upward arrows in the source table. This series reports TransClean Generation - Prompt 1 - Wrong Language from 2609.11399v1, S1.T1: Wrong-language Rate (%). Lower is better.
Metric: Wrong-language Rate (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 30B A3B 2507 (Thinking) | 0.51 | #384 |
| 2 | Qwen 3 30B A3B 2507 Instruct | 0.64 | #501 |
| 3 | Qwen 3 4B 2507 Instruct | 0.7 | #770 |
| 4 | Qwen 3 4B 2507 (Thinking) | 0.75 | #528 |
| 5 | DeepSeek R1 Distill Qwen 32B | 5.18 | #663 |
| 6 | Qwen 2.5 32B Instruct | 8.08 | #511 |
| 7 | aya-expanse-32B | 12.47 | #691 |
| 8 | Llama 3.2 3B Instruct | 15.87 | #1332 |
| 9 | Gemma 3 27B (IT) | 31.82 | #530 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=transclean-generation-prompt-1-wrong-language · How It Works · Data refreshed daily, snapshot 2026-11-07.