EVALITA - hate-speech-detection: leaderboard
The "hate-speech-detection" 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 | Llama 3.3 70B Instruct | 77.82 | #534 |
| 2 | Mistral Large 2 (Nov) Instruct (2411) | 77.03 | #465 |
| 3 | Qwen 2.5 72B Instruct | 76.38 | #454 |
| 4 | Gemma 3 27B (IT) | 75.87 | #523 |
| 5 | calme-3.2-instruct-78B | 75.82 | #347 |
| 6 | Llama-3.1-Tulu-3-70B | 75.51 | #482 |
| 7 | Qwen 2.5 VL 32B Instruct | 75.47 | #453 |
| 8 | DeepSeek R1 Distill Llama 70B | 74.6 | #594 |
| 9 | Gemma 2 27B (IT) | 74.14 | #645 |
| 10 | Mistral Small 3 | 72.9 | #627 |
| 11 | Qwen2.5-14B-Instruct-1M | 72.8 | #578 |
| 12 | Gemma 2 9B (IT) | 71.99 | #788 |
| 13 | Gemma 3 12B (IT) | 71.4 | #664 |
| 14 | LLaMAntino-3-ANITA-8B-Inst-DPO-ITA | 70.99 | #1009 |
| 15 | Qwen 3 Next 80B A3B Instruct | 69.84 | #368 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=evalita-hate-speech-detection · How It Works · Data refreshed daily, snapshot 2026-10-20.