Open Portuguese LLM - HateBR Offensive: leaderboard
Offensive-language detection task from the Open Portuguese LLM Leaderboard (HateBR), reporting macro-F1 on Brazilian Portuguese comments labelled offensive or non-offensive.
Metric: F1 (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 42 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 93.2 | #336 |
| 2 | GPT-5 Mini (2025-08-07) | 93.06 | #172 |
| 3 | Gemini 2.5 Pro (Preview 03-25) | 92.48 | #123 |
| 4 | GPT-5 (Minimal) | 92.33 | #98 (GPT-5) |
| 5 | DeepSeek V3.1 | 92.12 | #264 |
| 6 | GPT-5 Nano | 91.7 | #425 |
| 7 | Claude 3.7 Sonnet (20250219) | 91.25 | #201 |
| 8 | Gemini 1.5 Flash (001) | 90.92 | #577 |
| 9 | Llama 3.1 405B Instruct | 90.74 | #457 |
| 10 | DeepSeek V3 (0324) | 90.6 | #340 |
| 11 | Llama 4 Maverick Instruct | 90.48 | #456 |
| 12 | Gemini 1.5 Flash (002) | 90.46 | #583 |
| 13 | Llama 3.3 70B Instruct | 90.24 | #534 |
| 14 | Llama 3.2 90B Vision Instruct | 89.65 | #622 |
| 15 | Kimi K2 | 89.42 | #243 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-portuguese-llm-hatebr-offensive · How It Works · Data refreshed daily, snapshot 2026-10-20.