0

AIME 2025: Problems from the American Invitational Mathematics Examination

Saturated

A benchmark for evaluating AI's ability to solve challenging mathematics problems from the 2025 AIME - a prestigious high school mathematics competition.

Open
Publisher
Mathematical Association of America
Domain
Mathematics
License
mit
Published
Oct 2025
Notable for
Benchmark for evaluating Mathematics.
Canonical
github.com/UKGovernmentBEIS/inspect_evals/tree/main/src/inspect_evals/aime2025

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
AAprime-hub
Attribution policy →

Top score 98.7% by GPT-5 Codex (batch) - 204 models reporting (45 frontier)

Score history

204
0%25%50%75%100%Apr 24Sep 24Feb 25Jul 25Dec 25Phi-3 Mini Instruct 3.8BGPT-4o-miniPhi 4Grok 3 minio4 MiniGrok 4GPT-5 Codex (batch)

Top models

204
AIME 2025: Problems from the American Invitational Mathematics ExaminationBar chart with 21 bars. Highest value: GPT-5 Codex (batch) at 98.7.
21 models

Related tools

5
View all

Implementations, trainers, datasets and scaffolds linked to this eval.

AIME 2025 RL Env (Dev Team)

Dev Team

AIME-25 evaluation environment

ImplementationRL EnvMath

AIME 2025 RL Env (Prime Intellect)

Prime Intellect

AIME-25 evaluation environment

ImplementationRL EnvV1Math

AIME 2025 RL Env (Community)

AIME-25 evaluation environment

ImplementationRL EnvMath

VF Openbench RL Env (Community)

Environment for single-turn tasks in OpenBench

Trains towardRL Env

Deepconf RL Env (Community)

DeepConf environment for confidence-aware LLM reasoning evaluation

Trains towardRL EnvDeepconfConfidenceReasoning

FAQ

What is AIME 2025: Problems from the American Invitational Mathematics Examination?
A benchmark for evaluating AI's ability to solve challenging mathematics problems from the 2025 AIME - a prestigious high school mathematics competition.
What is the current top score on AIME 2025: Problems from the American Invitational Mathematics Examination?
The top reported score is 98.7% by GPT-5 Codex (batch), across 204 models reporting (45 from frontier labs).
How can a model improve its AIME 2025: Problems from the American Invitational Mathematics Examination score?
Tools linked to AIME 2025: Problems from the American Invitational Mathematics Examination on Sophon include AIME 2025 RL Env (Dev Team), AIME 2025 RL Env (Prime Intellect), AIME 2025 RL Env (Community), VF Openbench RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
What license is AIME 2025: Problems from the American Invitational Mathematics Examination under?
AIME 2025: Problems from the American Invitational Mathematics Examination is available under mit.