AIME 2025 (with tools)
Math · 2025-02-12
AIME 2025's 30 problems from the 2025 American Invitational Mathematics Examination solved with tools available, usually a Python interpreter or code execution; some reports do not itemize the tools. Tool runs approach a perfect score and are not comparable with runs without tools, which are scored under AIME 2025.
Benchmark source: Maa
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| Gemini 3 Pro Preview | 100.0 | Gemini 3 Flash model card · Google |
| Sonnet 4.5 | 100.0 | Gemini 3 Flash model card · Anthropic |
| GPT-5 Pro | 100.0 | GPT-5 launch post · OpenAI |
| Gemini 3 Flash Preview | 99.7 | Gemini 3 Flash model card · Google |
| GPT-5 | 99.6 | GPT-5 launch post · OpenAI |
| o4-mini | 99.5 | gpt-oss launch post · OpenAI |
| Nemotron 3 Nano 30B A3B | 99.2 | NVIDIA Nemotron 3 Nano technical report · NVIDIA |
| Kimi K2 Thinking | 99.1 | Kimi-K2-Thinking model card · Moonshot AI |
| gpt-oss-20b | 98.7 | gpt-oss launch post · OpenAI |
| Nemotron Cascade 2 30B A3B | 98.6 | Nemotron-Cascade-2-30B-A3B model card · NVIDIA |
| o3 | 98.4 | gpt-oss launch post · OpenAI |
| gpt-oss-120b | 97.9 | gpt-oss launch post · OpenAI |
| Gemini 2.5 Flash | 75.7 | Gemini 3 Flash model card · Google |
| Kimi K2 Instruct 0905 | 75.2 | Kimi-K2-Thinking model card · Moonshot AI |
| GPT-5 Chat | 71.0 | GPT-5 launch post · OpenAI |