BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

AIME (Vals)

Recent AIME problems, run by Vals AI at stated reasoning effort.

As of 25 Oct 2026, Gemini 3.1 Pro leads AIME (Vals) on BenchLeader with 98.1%, ahead of Muse Spark at 96.9%, across 92 model configurations with a published result.

Published by
Vals AI
Category
Maths
Index weight
0.5
Models
92
Data as of
25 Oct 2026
Newest result
16 Apr 2026

The publisher has posted no new results on this board since 16 Apr 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

Vals AI (vals.ai).

What the test looks like

Recent AIME competition papers, run by Vals AI with the model's reasoning effort recorded for each run. AIME answers are integers from 0 to 999.

How it is scored

Percent of problems answered correctly.

What to keep in mind

Thirty problems per year, so each is worth over three points. The papers are public soon after the exam, so newer models may have seen them.

Top 15
  1. 1Gemini 3.1 Pro (high)98.1%
  2. 2Muse Spark96.9%
  3. 3GPT-5.2 (xhigh)96.9%
  4. 4Gemini 3 Pro (high)96.7%
  5. 5GPT-5.4 (xhigh)96.7%
  6. 6Grok 4.20 (thinking)96.5%
  7. 7Claude Opus 4.796.3%
  8. 8Claude Opus 4.6 (thinking)95.6%
  9. 9Gemini 3 Flash (high)95.6%
  10. 10Kimi K2.5 (thinking)95.6%
  11. 11GPT-5.4 Mini (xhigh)95.6%
  12. 12Claude Opus 4.5 (thinking)95.4%
  13. 13Qwen3.6 Plus94.6%
  14. 14GPT-5 (high)93.4%
  15. 15GPT-5.1 (high)93.3%
92 of 92
#
1Gemini 3.1 Prohigh98.1%60.32026-04-16
2Muse Spark96.9%64.32026-04-16
3GPT-5.2xhigh96.9%60.72026-04-16
4Gemini 3 Prohigh96.7%60.02026-04-16
5GPT-5.4xhigh96.7%64.72026-04-16
6Grok 4.20thinking96.5%58.62026-04-16
7Claude Opus 4.796.3%63.22026-04-16
8Claude Opus 4.6thinking95.6%62.52026-04-16
9Gemini 3 Flashhigh95.6%58.32026-04-16
10Kimi K2.5thinking95.6%57.62026-04-16
11GPT-5.4 Minixhigh95.6%56.32026-04-16
12Claude Opus 4.5thinking95.4%59.92026-04-16
13Qwen3.6 Plus94.6%58.82026-04-16
14GPT-5high93.4%57.92026-04-16
15GPT-5.1high93.3%57.72026-04-16
16GLM 4.793.3%51.92026-04-16
17GLM-4.692.7%49.92026-04-16
18gpt-oss-120b92.6%45.62026-04-16
19Qwen3.5 Flash92.5%52.02026-04-16
20Claude Sonnet 4.692.3%57.12026-04-16
21GLM 5.191.9%55.92026-04-16
22Grok 4.1 Fastthinking91.9%54.72026-04-16
23GLM-5thinking91.7%58.92026-04-16
24GPT-5 minihigh91.5%54.12026-04-16
25Grok 4 Fastthinking91.3%51.42026-04-16
26MiniMax M2.791.0%55.62026-04-16
27Grok 490.6%58.02026-04-16
28MiniMax M2.588.8%53.62026-04-16
29GPT-5.4 Nanohigh88.8%48.42026-04-16
30Claude Sonnet 4.5thinking88.2%52.62026-04-16
31GLM-4.586.7%51.72026-04-16
32o3-minihigh86.5%47.22026-04-16
33Qwen3.5 Plusthinking86.0%57.32026-04-16
34gpt-oss-20b86.0%44.12026-04-16
35Gemini 2.5 Pro85.8%53.92026-04-16
36Kimi K2 Thinkingthinking85.4%52.42026-04-16
37o3high85.3%53.92026-04-16
38Grok 3 Mini Fasthigh85.0%53.72026-04-16
39DeepSeek V3.2high84.6%–2026-04-16
40Qwen3 235B A22B84.0%50.12026-04-16
41o4-minihigh83.7%50.12026-04-16
42Magistral Medium 1.283.5%43.92026-04-16
43Gemini 3.1 Flash Litehigh83.3%48.82026-04-16
44Claude Haiku 4.5thinking82.7%46.72026-04-16
45GPT-5 nanohigh81.2%47.42026-04-16
46Qwen3 Maxmax81.0%52.82026-04-16
47Magistral Small 1.280.7%40.32026-04-16
48Claude Opus 4.1thinking78.2%56.42026-04-16
49MiniMax-M2.177.9%53.82026-04-16
50Claude Opus 4.576.9%57.92026-04-16
51Claude Sonnet 4thinking76.3%53.22026-04-16
52DeepSeek R174.0%46.82026-04-16
53o1high71.5%48.32026-04-16
54Grok 3 Mini Fastlow70.6%50.72026-04-16
55Kimi K262.7%50.42026-04-16
56Grok 358.8%51.52026-04-16
57Langston Nim Nvidia Llama 3.3 Nemotron Super 49B v1 42e84561thinking53.5%–2026-04-16
58Gemini 2.5 Flash 09 2025thinking51.5%51.12026-04-16
59Gemini 2.5 Flash 09 202549.8%46.92026-04-16
60GPT-4.1 minihigh49.4%46.32026-04-16

More maths benchmarks · All LLM benchmarks

Cite as: BenchLeader, “AIME (Vals) leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/vals_aime.

AIME (Vals): questions

What does AIME (Vals) measure?
Recent AIME competition papers, run by Vals AI with the model's reasoning effort recorded for each run. AIME answers are integers from 0 to 999. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads AIME (Vals)?
Gemini 3.1 Pro leads AIME (Vals) with 98.1% as of 25 Oct 2026, ahead of Muse Spark at 96.9%.
How many models have AIME (Vals) results?
92 model configurations have a AIME (Vals) result on BenchLeader, all taken from Vals AI.
Who runs AIME (Vals) and how often is it updated?
AIME (Vals) is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does AIME (Vals) count toward the BenchLeader Index?
Yes. AIME (Vals) contributes to the maths category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.