Benchmarks / Reasoning & Math

AIME

American Invitational Mathematics Examination problems (recent years).

maintainer unknown · 269 observations on record

Versions

1 version

Version	Tasks	Window	Released	Notes
AIME current aime	—	—	—

Current leaderboard

AIME · Accuracy · top 25 of 214 systems

#	System	Accuracy %	Verification	Source
1	GPT 5.2 (2025-12-11)xhigh effort OpenAI	99.0%	independent	Artificial Analysis
2	GPT 5 Codexhigh effort OpenAI	98.7%	independent	Artificial Analysis
3	Gemini 3.1 Pro Previewhigh effort Google DeepMind	98.1%	independent	Vals AI
4	Gemini 3 Flash Previewthinking Google DeepMind	97.0%	independent	Artificial Analysis
5	Muse Spark Meta AI	96.9%	independent	Vals AI
6	Gemini 3 Pro Previewhigh effort Google DeepMind	96.7%	independent	Vals AI
7	GPT 5.2 (2025-12-11)medium effort OpenAI	96.7%	independent	Artificial Analysis
8	GPT 5.4 (2026-03-05)xhigh effort OpenAI	96.7%	independent	Vals AI
9	Grok 4.20 0309 Reasoning xAI	96.5%	independent	Vals AI
10	Mimo V2 Flashthinking xiaomi	96.3%	independent	Artificial Analysis
11	Claude Opus 4.7 Anthropic	96.3%	independent	Vals AI
12	GPT 5.1 Codexhigh effort Unknown	95.7%	independent	Artificial Analysis
13	GPT 5 (2025-08-07)high effort OpenAI	95.7%	independent	Artificial Analysis
14	GPT 5.4 Mini (2026-03-17)xhigh effort OpenAI	95.6%	independent	Vals AI
15	Kimi K2 5thinking Moonshot AI	95.6%	independent	Vals AI
16	Claude Opus 4.6thinking Anthropic	95.6%	independent	Vals AI
17	Gemini 3 Flash Previewhigh effort Google DeepMind	95.6%	independent	Vals AI
18	Claude Opus 4.5 20251101thinking Anthropic	95.4%	independent	Vals AI
19	GLM 4.7thinking Z.ai	95.0%	independent	Artificial Analysis
20	Kimi K2 Thinking Unknown	94.7%	independent	Artificial Analysis
21	Qwen3 6 Plus Alibaba	94.6%	independent	Vals AI
22	Grok 4 xAI	94.3%	independent	Artificial Analysis
23	Qwen3 235B A22b 2507thinking Alibaba	94.0%	independent	Artificial Analysis
24	GPT 5.1 (2025-11-13)high effort OpenAI	94.0%	independent	Artificial Analysis
25	O4 Mini (2025-04-16)high effort OpenAI	94.0%	independent	Artificial Analysis

One row per evaluated system — reasoning-effort variants rank separately. Where a system has attr variants for this version (eval splits, style control), only the canonical variant is ranked: split=semi_private, style_control=true, or the unsplit run.

Source agreement

where evaluating organizations agree — and don't

AIMEArtificial Analysis vs Vals AI

overlap 27 systemsrank correlation 0.92mean |Δ| 3.25max |Δ| 16.09

most contested: Kimi K2 (-16.09) · Qwen3 Max Preview (+14.31) · O4 Mini (2025-04-16) (+10.33)

Possible reasons: run setting 'implementation' differs (artificial-analysis vs vals-ai).