Open Portuguese LLM - BLUEX: leaderboard
Multiple-choice questions from the entrance exams of UNICAMP and USP, two Brazilian universities. From the final table of the archived Open Portuguese LLM Leaderboard (3-shot, lm-evaluation-harness), covering about 1,210 open checkpoints and 42 API models; accuracy (%), higher is better. Frozen since the last evaluation on 2025-09-01.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 1253 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 03-25) | 94.99 | #126 |
| 2 | GPT-5 Mini (2025-08-07) | 91.38 | #175 |
| 3 | Gemini 2.5 Flash | 86.51 | #255 |
| 4 | GPT-5 Nano | 85.26 | #439 |
| 5 | Claude 3.7 Sonnet (20250219) | 84.56 | #209 |
| 6 | DeepSeek V3 (0324) | 84.14 | #349 |
| 7 | Qwen 3 14B | 83.31 | #542 |
| 8 | Qwen 3 30B A3B | 83.03 | #496 |
| 9 | Kimi K2 | 82.75 | #252 |
| 10 | DeepSeek V3.1 | 81.78 | #275 |
| 11 | Llama 4 Maverick Instruct | 81.22 | #467 |
| 12 | Qwen 3 8B | 80.67 | #678 |
| 13 | Qwen 2.5 VL 72B Instruct | 80.53 | #383 |
| 14 | Qwen 2.5 VL 32B Instruct | 80.53 | #463 |
| 15 | Gemini 2.0 Flash (001) | 80.39 | #365 |
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-bluex · How It Works · Data refreshed daily, snapshot 2026-11-05.