The honest view from inside technical leadership
LLM Leaderboard · Reasoning

ARC-AGI-2 leaderboard

Novel visual puzzles that require inferring an abstract rule from a few examples. Each task is new, so memorised knowledge does not help.

← All models

Rankings

Snapshot 22 November 2026 · Scores copied from each benchmark's own leaderboard · How the CTO Index works

Score and price

#ModelProviderARC-AGI-2Price $/M
1GPT-6.2 AstraOpenAI95.8%$8 / $40
2GPT-6 AstraOpenAI95.0%$10 / $50
3Gemini 4 CarbonGoogle94.2%$2 / $10
4GPT-6.1 SolOpenAI94.2%$2 / $10
5Claude Fable 5.2Anthropic93.8%$8 / $40
6Claude Opus 5.5Anthropic93.3%$4 / $20
7Gemini 4 HydrogenGoogle92.5%$1.50 / $7.50
8GPT-5.6 SolOpenAI92.5%$4 / $20
9Gemini 4 ArgonGoogle91.3%$2 / $10
10Claude Opus 5Anthropic90.4%$5 / $25
11Claude Fable 5.1Anthropic90.0%$10 / $50
12GPT-6 SolOpenAI89.6%$2 / $10
13Claude Fable 5Anthropic89.2%$10 / $50
14Gemini 3.8 FlashGoogle89.2%$0.75 / $3.75
15GPT-6.2 SolOpenAI85.8%$2 / $10
16Gemini 3.7 FlashGoogle84.6%$0.75 / $3.75
17Claude Sonnet 5.5Anthropic83.9%$2 / $10
18Inkling 2 SmallThinking Machines81.7%$0.30 / $1
19DeepSeek V4.1 FlashDeepSeek78.1%$0.15 / $0.60
20Claude Opus 4.8Anthropic72.1%$5 / $25
21Grok 4.6xAI67.1%$1.60 / $4.80
22GLM-5.3-FlashZ.ai65.8%$0.02 / $0.07
23Grok 4.7xAI61.4%$2 / $6
24Claude Haiku 5.5Anthropic61.3%$0.10 / $0.50
25Kimi K3Moonshot AI60.4%$3 / $15
26GPT-6 LunaOpenAI59.3%$0.10 / $0.50
27Qwen3.8-27BAlibaba42.4%$0.40 / $3
28Beam 1.0Reflection AI23.5%$0.30 / $1

Other benchmarks

Frequently asked questions

What does ARC-AGI-2 measure?

Novel visual puzzles that require inferring an abstract rule from a few examples. Each task is new, so memorised knowledge does not help.

Which model has the highest ARC-AGI-2 score?

GPT-6.2 Astra from OpenAI leads ARC-AGI-2 at 95.8%, ahead of GPT-6 Astra (95.0%).

Why does ARC-AGI-2 matter to engineering leaders?

A signal for how well a model handles problems it has not seen before, such as an unfamiliar codebase, an odd incident or a design trade-off with no template.

What are the limitations of ARC-AGI-2?

Top models now score above 90%, so the benchmark is near saturation at the frontier. Cost per task varies by more than 10x between models with similar scores.