Math-PT (pt-BR OE) - Level 3: leaderboard

Math-PT (2026) tests mathematical reasoning on problems written natively in Brazilian Portuguese; this board covers the level-3 open-ended questions from OBM. Accuracy in percent, with Kimi K2 Thinking checking that the final answer is mathematically equivalent to the gold answer, higher is better.

Metric: Accuracy (%) on the 58 level-3 open-ended questions in Brazilian Portuguese (OBM); native-language zero-shot prompt asking for a boxed final answer, figures given as LaTeX source; Kimi K2 Thinking judges whether the final answer is mathematically equivalent to the gold answer; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 235B A22B93.1#628
2GPT-591.38#178
3Gemini 2.5 Flash91.38#463
4DeepSeek V3.186.21#511
5Qwen 3 14B84.48#1269
6Qwen 3 32B82.76#972
7Qwen 3 30B A3B74.14#1156
8Qwen 3 8B72.41#1873
9Claude Haiku 4.570.69#583
10Gemma 3 27B (IT)67.24#1324
11Llama 3.3 70B Instruct34.48#1385

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

Interactive version: theaggregate.ai/benchmark?slug=math-pt-pt-br-oe-level-3 · How It Works · Data refreshed daily, snapshot 2026-11-21.