Skip to content
Leaderboards.

MMLU-Pro leaderboard: who is #1

TIGER-Lab · Accuracy, higher is better · 100 models · stale since Mar 11, 2026, not counted

TIGER-Lab

Top 25 of 100 models, best variant per model as published

#ModelScoreCorrect
1Gemini 3.1 Pro Previewas “Gemini-3.1-Pro”91.2%
2Gemini 3 Proas “Gemini-3-Pro(11/25)”90.1%
3o1as “GPT-o1”89.3%
4Claude Opus 4.6as “Claude-4.6-Opus(Thinking)”89.1%
5Gemini 3 Flashas “Gemini-3-Flash(12/25)”88.6%
6MiniMax M2.1as “MiniMax-M2.1”88.0%
7Qwen3.5 397B A17Bas “Qwen3.5-397B-A17B”87.8%
8Seed2.0 Liteas “Seed2.0-Lite”87.7%
9GPT-5.4OpenAI87.5%
10Claude Sonnet 4.5as “Claude-4.5-Sonnet(Thinking)”87.4%
10GPT-5.2OpenAI87.4%
12Claude Opus 4as “Claude-4-Opus-Thinking”87.3%
12Claude Opus 4.5as “Claude-4.5-Opus(Thinking)”87.3%
12Claude Sonnet 4.6as “Claude-4.6-Sonnet(Thinking)”87.3%
15Hunyuan T1as “Hunyuan-T1”87.2%
16GPT-5as “GPT-5(high)”87.1%
16K2.5 1t A32Bas “K2.5-1T-A32B”87.1%
18Grok 4as “Grok-4”87.0%
18Seed V1.5as “Seed-Thinking-v1.5”87.0%
18Seed2.0 Proas “Seed2.0-Pro”87.0%
21Qwen3.5 122B A10Bas “Qwen3.5-122B-A10B”86.7%
22Seed1.6as “Seed1.6-Thinking”86.6%
22Seed1.6 Baseas “Seed1.6-Base”86.6%
24GPT-5.1OpenAI86.4%
24Seed1.6 Adaas “Seed1.6-Ada-Thinking”86.4%
About this board and its history

What it measures

12,000 harder, ten-option multiple-choice questions across 14 academic subjects.

Other reasoning boards

Source

Scores as published by TIGER-Lab on the capture date (2026-09-24). Source: TIGER-Lab results.csv on Hugging Face.

huggingface.co/spaces/TIGER-Lab/MMLU-Pro

Sources: each leaderboard's own published results page, captured every day at 04:30 UTC. Scores exactly as each board published them; we run no evaluations. Logos via logo.dev; trademarks belong to their owners.

New #1s by email

Saturdays, only in weeks when a leaderboard has a new #1.

Double opt-in. Unsubscribe any time.