DiBiMT: leaderboard
Divide and Benchmark in Machine Translation: word sense disambiguation benchmark for MT systems across multiple language pairs. Its home at nlp.uniroma1.it/dibimt has been unreachable since 2026-08-19, the whole Sapienza NLP host refuses connections: so this card points at the newest Wayback capture of the board, from 2025-08-13.
Metric: Avg Accuracy (%). Source: web.archive.org. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 71.09 | #438 |
| 2 | Gemma 2 9B | 69.8 | #958 |
| 3 | GPT-3.5 Turbo | 67.3 | #865 |
| 4 | Llama 3 8B | 55.45 | #1242 |
| 5 | Mistral 7B | 48.31 | #1397 |
| 6 | gemma-7B | 46.79 | #1384 |
| 7 | Llama 2 7B | 46.05 | #1520 |
| 8 | gemma-2B | 40.53 | #1575 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=dibimt · How It Works · Data refreshed daily, snapshot 2026-10-20.