📐 Ranking: Matemática

De olimpíadas de matemática a problemas de fronteira: AIME, MATH e FrontierMath.

Modelos: 32

#	Modelo	Podium Score	AIME 2025	FrontierMath	MATH	Velocidade
01	Claude Mythos Preview Anthropic · preview	100.0	—	—	—	80 t/s
02	MiMo V2.5 Pro Xiaomi · Aberto	100.0	—	—	86.2%	65 t/s
03	Grok 4 Heavy xAI · preview	84.6	100%	—	—	50 t/s
4	Qwen3.7 Max Alibaba	81.6	—	—	—	203 t/s
5	Claude Opus 4.6 Anthropic	79.6	99.8%	—	—	50 t/s
6	Gemini 3.1 Pro Google	76.0	—	—	—	136 t/s
7	Gemini 3 Pro Google	72.9	100%	—	—	120 t/s
8	Kimi K3 Moonshot AI · Aberto	70.9	—	—	—	37 t/s
9	GPT-5.6 Sol OpenAI	70.7	—	89%	—	72 t/s
10	GLM-5.2 Z.ai · Aberto	69.8	—	—	—	164 t/s
11	Gemini 3 Flash Google	69.6	99.7%	—	—	120 t/s
12	DeepSeek V4 Pro DeepSeek · Aberto	66.5	—	—	—	66 t/s
13	Qwen3.7 Plus Alibaba	62.0	—	—	—	54 t/s
14	DeepSeek V4 Flash DeepSeek · Aberto	62.0	—	—	—	113 t/s
15	Muse Spark Meta	59.8	—	—	—	80 t/s
16	Qwen3.6 Plus Alibaba	59.2	—	—	—	55 t/s
17	Claude Opus 4.8 Anthropic	58.7	—	—	—	50 t/s
18	GPT-5.6 Terra OpenAI	57.3	—	84.9%	—	127 t/s
19	Claude Opus 4.7 Anthropic	57.0	—	—	—	80 t/s
20	GPT-5.2 OpenAI	53.0	100%	40.3%	—	90 t/s
21	Seed 2.0 Pro ByteDance	47.2	98.3%	—	—	40 t/s
22	Kimi K2.6 Moonshot AI · Aberto	46.9	—	—	—	40 t/s
23	Qwen3.5 397B-A17B Alibaba · Aberto	45.3	—	—	—	71 t/s
24	GPT-5.6 Luna OpenAI	40.3	—	78.6%	—	175 t/s
25	Hunyuan Hy3 Tencent	35.8	—	—	—	69 t/s
26	GPT-5.4 OpenAI	32.1	—	47.6%	—	60 t/s
27	Claude Sonnet 4.6 Anthropic	30.7	—	—	—	50 t/s
28	GPT-5.5 OpenAI	25.7	—	35.4%	—	60 t/s
29	GPT-5.1 OpenAI	13.0	94%	—	—	60 t/s
30	GPT-5 OpenAI	10.9	94.6%	—	84.7%	60 t/s
31	Qwen3.8 Max Alibaba	5.0	—	—	—	55 t/s
32	DeepSeek V3.2 DeepSeek · Aberto	0.0	93.1%	—	—	60 t/s

O Podium Score é uma mistura ponderada de cinco leaderboards. Saiba mais na metodologia.