BLUEX v2 (Official): leaderboard

BLUEX v2 evaluates open-ended Portuguese university entrance-exam answers. The official full results include 21 models over 880 evaluated subquestions, scored from 0 to 10 by a Sabia-4 rubric judge, with evaluation errors counted as zero. All models receive image captions, including text-only models. The judge also appears among the subjects. This is the full roster of the same experiment whose paper prints only the top and bottom five models; higher is better.

Metric: Mean rubric score (0-10; 880 subquestions; evaluation errors score zero). Source: github.com. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 21 models tracked. Updated 2 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)9.1#99
2GPT-5.48.92#135
3GPT-5 Mini8.9#359
4Qwen 3.5 122B A10B8.83#341
5Gemini 3.1 Flash Lite (Preview)8.83#418
6Qwen 3.5 35B A3B8.77#490
7Qwen 3.5 27B8.74#374
8Qwen 3.5 9B8.63#822
9Sabia-48.6#443
10Gemma 4 26B A4B (IT)8.6#561
11Sabiazinho-48.58#666
12Qwen 3 14B8.57#1264
13Qwen 3 30B A3B 2507 Instruct8.54#1091
14Mistral Small 48.43#1232
15Qwen 3 8B8.41#1868

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=bluex-v2-official · How It Works · Data refreshed daily, snapshot 2026-11-17.