AIME 2025: what the benchmark actually measures
A practical guide to aime-2025, using data from the AI & LLM Index. The goal is a shortlist and a decision, not a tour of every option.
What AIME 2025 measures
AIME 2025 is a mathematics benchmark. Problems from the 2025 American Invitational Mathematics Examination.
The index scores all 600 models on a 0–100 scale. The average is 63.6 and the top score is 91.4.
How to read the scores
A benchmark score is a compressed signal. It tells you how a model performed on a specific test distribution, not how it will perform on your data. Read the leaderboard as a way to form a shortlist, then validate the shortlist with your own examples.
On AIME 2025, the gap between the leader and the median is 27.8 points. That gap is the useful part of the chart: it shows how much separation the benchmark creates.
Leaders
Claude Opus 4.5 Pro leads at 91.4, followed by GPT-5.2 Pro at 91.0 and GPT-5.1 Pro at 90.6.
Limitations
- Scores in this index are modelled estimates, not official results.
- Benchmark contamination can inflate scores for models trained on similar data.
- A high score does not guarantee good performance on a different task shape.
- Prompt format and evaluation harness can move results by several points.
Top 5 on AIME 2025
| Model | Provider | Score | Input / 1M |
|---|---|---|---|
| Claude Opus 4.5 Pro | Anthropic | 91.4 | $20.98 |
| GPT-5.2 Pro | OpenAI | 91.0 | $21.32 |
| GPT-5.1 Pro | OpenAI | 90.6 | $19.77 |
| GPT-5.2 | OpenAI | 90.6 | $19.71 |
| o4 Reasoning | OpenAI | 90.3 | $19.69 |
Related
FAQ
Who leads AIME 2025?
Claude Opus 4.5 Pro leads this index at 91.4.
What is the most important factor in aime-2025?
Fit to the task beats raw benchmark rank. Define the failure modes, pick the benchmarks that resemble them, then test a small shortlist on your own examples.
How many models should I evaluate?
Three is usually enough: a quality candidate, a value candidate and a fallback. More candidates add work without changing the decision.
How do I control cost?
Compare blended 3:1 price, watch output tokens, use caching where available and route simple requests to a cheaper model.