EuroEval Portuguese NLU - HAREM: leaderboard
The "HAREM" subtask of EuroEval Portuguese NLU. Portuguese-language EuroEval natural-language-understanding suite, separating NLU task performance from the broader all-task EuroEval aggregate.
Metric: Micro F1 excluding MISC (%). Source: euroeval.com. Saturation forecast: Around 2029. 438 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B 2507 FP8 (Thinking) | 64.6 | #246 (Qwen 3 235B A22B 2507 FP8) |
| 2 | Gemma 4 31B (IT) | 63.85 | #176 |
| 3 | Gemma 4 26B A4B (IT) | 63.12 | #297 |
| 4 | Qwen 3 4B 2507 (Thinking) | 62.2 | #538 (Qwen 3 4B 2507) |
| 5 | Claude Sonnet 4.5 (Thinking) | 61.29 | #142 (Claude Sonnet 4.5) |
| 6 | Voxtral-Small-24B-2507 | 61.2 | #581 |
| 7 | Llama 3.1 70B | 61.1 | #592 |
| 8 | Qwen 3 Next 80B A3B (Thinking) | 60.81 | #320 (Qwen 3 Next 80B A3B) |
| 9 | Mistral Small 3.1 | 60.73 | #610 |
| 10 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 60.47 | #679 |
| 11 | OLMo 3.1 32B (Thinking) | 60.41 | #632 (OLMo 3.1 32B) |
| 12 | Llama 3 70B | 59.36 | #631 |
| 13 | Mistral Small 3.2 | 59.09 | #605 |
| 14 | GLM 4.5 Air (Thinking) | 58.3 | #392 (GLM 4.5 Air) |
| 15 | QwQ-32B | 58.1 | #421 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=euroeval-portuguese-nlu-harem · How It Works · Data refreshed daily, snapshot 2026-10-20.