0

MathArena

Frontier

Live leaderboard of LLMs on recent math-olympiad and research-style problems, refreshed monthly to minimise pretraining contamination.

Domain
math
Published
May 2026
Updates
Monthly
Notable for
Continuously refreshed math benchmark — pulls fresh problems each month so frontier models can't game contamination, and reports per-problem accuracy with confidence intervals.
Canonical
matharena.ai
Official leaderboard
matharena.ai

Cite

Notes

Only stored in your browser.

Attribution

Leaderboard scores
matharena
Attribution policy →

Top score 92.8% by GPT-5.5 - 23 models reporting (9 frontier)

Score history

22
40%55%70%85%100%Jul 25Oct 25Jan 26Apr 26Qwen3 30B A3B Instruct 2507Gemini 3 ProGPT-5.2Claude Opus 4.6GPT-5.5

Top models

23
MathArenaBar chart with 21 bars. Highest value: GPT-5.5 at 92.8.
21 models

Where it's ranked

1

Related tools

1
View all

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

FAQ

What is MathArena?
Live leaderboard of LLMs on recent math-olympiad and research-style problems, refreshed monthly to minimise pretraining contamination.
What is the current top score on MathArena?
The top reported score is 92.8% by GPT-5.5, across 23 models reporting (9 from frontier labs).
How can a model improve its MathArena score?
Tools linked to MathArena on Sophon include APEX Shortlist RL Env (Prime Intellect) - RL environments, datasets, and scaffolds that target this eval.