EVALITA - lexical-substitution: leaderboard
The "lexical-substitution" subtask of EVALITA. A task from EVALITA, the periodic evaluation campaign for Italian-language natural language processing and understanding.
Metric: CPS. Source: huggingface.co. Saturation forecast: Around January 2027. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 Next 80B A3B Instruct | 54.53 | #368 |
| 2 | Mistral Small 3 | 52.67 | #627 |
| 3 | Qwen 2.5 72B Instruct | 50.24 | #454 |
| 4 | Llama 3.3 70B Instruct | 46.8 | #534 |
| 5 | Llama-3.1-Tulu-3-70B | 46.45 | #482 |
| 6 | Gemma 3 27B (IT) | 42.41 | #523 |
| 7 | Lexora-Medium-7B | 40.88 | #773 |
| 8 | Qwen 2.5 VL 32B Instruct | 40.67 | #453 |
| 9 | Llama 4 Scout Instruct | 39.29 | #539 |
| 10 | calme-3.2-instruct-78B | 38.82 | #347 |
| 11 | DeepSeek R1 Distill Llama 70B | 38.61 | #594 |
| 12 | Mistral Large 2 (Nov) Instruct (2411) | 38.56 | #465 |
| 13 | Qwen2.5-14B-Instruct-1M | 36.76 | #578 |
| 14 | Gemma 2 27B (IT) | 36.04 | #645 |
| 15 | Phi-3-medium-4k-instruct | 35.87 | #952 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=evalita-lexical-substitution · How It Works · Data refreshed daily, snapshot 2026-10-20.