What is AIME?
Problems from the American Invitational Mathematics Examination, a competition-level math benchmark used to test multi-step quantitative reasoning.
AIME scores by model
2
GPT-5.2 ProOpenAI100% ↗
3
Dola Seed 2.0 ProByteDance Seed99% ↗
4
Step 3.5 FlashStepFun98.3% ↗
5
MAI-Thinking-1Microsoft97% ↗
7
Gemini 3 Flash PreviewGoogle95.2% ↗
10
DeepSeek V3.2DeepSeek94.2% ↗
12
Grok 4 Fast ReasoningxAI92% ↗
14
Amazon Nova 2 LiteAmazon91% ↗
15
NVIDIA Nemotron 3 Super 120B-A12BNVIDIA90.21% ↗
16
gpt-oss-120bOpenAI90% ↗
17
Command A+Cohere90% ↗
18
gpt-oss-20bOpenAI89.2% ↗
20
NVIDIA Nemotron 3 Nano 30B-A3B BF16NVIDIA89.1% ↗
22
Gemini 2.5 ProGoogle88.3% ↗
23
GPT-5 miniOpenAI87.5% ↗
24
GPT-5 nanoOpenAI85% ↗
25
Ministral 3 14BMistral AI85% ↗
26
Claude Sonnet 4.5Anthropic84.2% ↗
27
GLM-4.5-AirZ.ai83.3% ↗
28
Qwen3 235B-A22BQwen80.8% ↗
29
Ministral 3 8BMistral AI78.7% ↗
30
Ministral 3 3BMistral AI72.1% ↗
31
Gemini 2.5 FlashGoogle72% ↗
32
DeepSeek R1DeepSeek70% ↗
33
Gemini 2.5 Flash-LiteGoogle63.1% ↗
34
Phi-4 ReasoningMicrosoft62.9% ↗
35
LongCat Flash ChatMeituan LongCat61.25% ↗
36
Command A ReasoningCohere57% ↗
37
Claude 3.7 SonnetAnthropic49.2% ↗
38
DeepSeek V3DeepSeek25% ↗
40
Claude 3.5 SonnetAnthropic3.3% ↗