UnifyBench

RankingARC-AGI-2

ARC-AGI-2

Data updated 22 Oct 2026

Bucket
Hard reasoning
Unit
percent
Direction
Higher is better
Version
2
Display harness
ARC Prize verified
Board
https://arcprize.org/leaderboard

Compare published benchmark results with category weights →

Models

Results are ordered by score in the display harness, followed by models with no result. Other harnesses are listed separately below. Missing results remain unknown.

26–50 of 471 entries

ModelScoreEvidenceSource-recorded date
DeepSeek V4-Pro 0813DeepSeek61.3%official board2026-08-13
Claude Sonnet 4.6Anthropic60.4%official board2026-02-17
Gemini 3.6 FlashGoogle60.4%official board2026-07-21
Kimi K3Moonshot60.4%official board2026-07-16
GPT-5.6 LunaOpenAI59.5%official board2026-07-09
GPT-6 LunaOpenAI59.3%official board2026-09-22
GPT-5.2 ProOpenAI54.2%official board2025-12-11
GPT-5.2OpenAI52.9%official board2025-12-11
Qwen3.8-27BQwen42.4%official board2026-08-14
Gemini 3 Flash PreviewGoogle33.6%official board2025-12-17
GLM-5.2Z.ai22.8%official board2026-06-13
GPT-5.4 MiniOpenAI18.9%official board2026-03-17
GPT-5 ProOpenAI18.3%official board2025-10-06
Gemini 3.5 Flash-LiteGoogle10.3%official board2026-07-21
GPT-5OpenAI9.9%official board2025-08-07
o3OpenAI6.5%official board2025-04-16
o4-miniOpenAI6.1%official board2025-04-16
GPT-5.4 NanoOpenAI5.7%official board2026-03-17
GLM-5Z.ai4.9%official board2026-02-20
MiniMax-M2.5MiniMax4.9%official board2026-02-12
o3-proOpenAI4.9%official board2025-06-10
GPT-5 MiniOpenAI4.4%official board2025-08-07
DeepSeek-V3.2DeepSeek4%official board2025-12-01
Claude Sonnet 4.5Anthropic3.8%official board2025-09-29
o3-miniOpenAI3%official board2025-01-31