ARC-AGI-2 benchmark

61 AI models ranked on ARC-AGI-2. Each row shows the model’s score next to its cheapest API price per 1M tokens where it has one, so you can weigh quality against cost.

Top model today

GPT-5.6 Sol (OpenAI) leads at 93%, from $5.00 per 1M input.

Data via ARC Prize

Models ranked

61

on this benchmark

Top score

93%

current leader

Type

Single eval

benchmark

Updated

15 July 2026

last source fetch

Cost vs ARC-AGI-2

55 priced models
Most attractive quadrant (cheap + high)
= reasoning model (extended thinking)

Each labelled name links to that model’s page. Curated to the cost-efficiency frontier, top scorers and cheapest so labels stay readable; the full field is in the leaderboard table below (and via “select all”). Reasoning models (extended “thinking” before answering) carry a ring around the dot.

Models ranked on ARC-AGI-2

highest score first · 61 models
Maker
#
1GPT-5.6 SolOpenAI93%max$1.44 / task$5.00$30.00
2GPT-5.5OpenAI85%xhigh$1.87 / task$5.00$30.00
3GPT-5.5 ProOpenAI85%high$10.51 / taskn/an/a
4GPT-5.6 TerraOpenAI84%max$1.09 / task$2.50$15.00
5GPT-5.4 ProOpenAI83%xhigh$16.41 / taskn/an/a
6Gemini 3.1 Pro PreviewGoogle77%$2.00$12.00
7Claude Opus 4.7Anthropic76%max$7.43 / task$5.00$25.00
8GPT-5.4OpenAI74%xhigh$1.52 / task$2.50$15.00
9Claude Opus 4.8Anthropic72%high$2.74 / task$5.00$25.00
10Gemini 3.5 FlashGoogle72%high$0.850 / task$1.50$9.00
11Claude Opus 4.6Anthropic69%high$3.47 / task$5.00$25.00
12Grok 4 20 Beta 0309 ReasoningxAI65%$2.00$6.00
13Claude Sonnet 4.6Anthropic60%high$2.70 / task$3.00$15.00
14GPT-5.6 LunaOpenAI60%max$0.670 / task$1.00$6.00
15GPT-5.2 ProOpenAI54%high$15.72 / task$21.00$168.00
16GPT 5.2 (2025-12-11)OpenAI53%xhigh$1.90 / task$1.75$14.00
17Gemini 3 Deep ThinkGoogle45%n/an/a
18Claude Opus 4.5 (2025-11-01)Anthropic38%reasoning: true · thinking_budget: 64000$2.40 / task$5.00$25.00
19Gemini 3 Flash PreviewGoogle34%high$0.231 / task$0.50$3.00
20Gemini 3 Pro PreviewGoogle31%$2.00$12.00
21GLM 5.2Zhipu23%$1.40$4.40
22GPT-5.4 MiniOpenAI19%xhigh$0.750 / task$0.75$4.50
23GPT 5.1 (2025-11-13)OpenAI18%high$1.17 / task$1.25$10.00
24Grok 4 0709xAI16%$3.00$15.00
25Claude Sonnet 4.5 (2025-09-29)Anthropic14%reasoning: true · thinking_budget: 32000$0.759 / task$3.00$15.00
26Kimi K2.5Moonshot12%$0.50$2.80
27GPT 5 (2025-08-07)OpenAI10%high$0.730 / task$1.25$10.00
28Claude Opus 4 (2025-05-14)Anthropic9%reasoning: true · thinking_budget: 16000$1.93 / task$15.00$75.00
29O3 (2025-04-16)OpenAI7%high$0.834 / task$2.00$8.00
30O4 Mini (2025-04-16)OpenAI6%high$0.856 / task$1.10$4.40
31Claude Sonnet 4 (2025-05-14)Anthropic6%reasoning: true · thinking_budget: 16000$0.486 / task$3.00$15.00
32GPT-5.4 NanoOpenAI6%xhigh$0.160 / task$0.20$1.25
33Grok 4 Fast ReasoningxAI5%$0.20$0.50
34Gemini 2.5 ProGoogle5%reasoning: true · thinking_budget: 32000$0.757 / task$1.25$10.00
35GLM 5Zhipu5%$0.80$2.56
36MiniMax M2.5MiniMax5%n/an/a
37O3 ProOpenAI5%high$7.55 / taskn/an/a
38GPT 5 Mini (2025-08-07)OpenAI4%high$0.198 / task$0.25$2.00
39DeepSeek V3.2DeepSeek4%$0.27$0.40
40Claude Haiku 4.5 (2025-10-01)Anthropic4%reasoning: true · thinking_budget: 32000$0.377 / task$1.00$5.00
41O3 Mini (2025-01-31)OpenAI3%high$0.547 / task$1.10$4.40
42GPT 5 Nano (2025-08-07)OpenAI3%high$0.029 / task$0.05$0.40
43Gemini 2.0 FlashGoogle1%$0.10$0.40
44R1DeepSeek1%$1.35$5.40
45Qwen3 235B A22b Instruct 2507Alibaba1%$0.09$0.58
46R1 0528DeepSeek1%$0.20$0.25
47Claude 3.7 SonnetAnthropic1%reasoning: true · thinking_budget: 8000$0.360 / task$3.00$15.00
48O1 Mini (2024-09-12)OpenAI1%$1.10$4.40
49GPT 4.5 PreviewOpenAI1%$75.00$150.00
50Gemini 1.5 Pro (Sep)Google1%n/an/a
150 of 61
Page 1 of 2
Prices are on-demand list rates. Enterprise commitments can be materially lower. Cost per task uses best-effort token usage.

Frequently asked questions

What is the ARC-AGI-2 benchmark?

ARC-AGI-2 is the second-generation Abstraction and Reasoning Corpus from ARC Prize, a test of fluid intelligence rather than recalled knowledge. Each task shows a few input-output grid examples and the model must infer the underlying transformation and apply it to a new grid. The tasks resist memorisation, so frontier models still score far below the human panel. Scored as the percentage of tasks solved on the semi-private evaluation set, which ARC Prize reports alongside each system's cost per task.

Which AI model scores highest on ARC-AGI-2?

GPT-5.6 Sol (OpenAI) leads with 93%, from $5.00 per 1M input tokens.

How many models are ranked on ARC-AGI-2?

61 models carry a ARC-AGI-2 score. Scores come from ARC Prize; where a model has a live API price it is verified against LiteLLM and OpenRouter, and models with no current price are listed without one.

Other benchmarks

compare the same models on a different eval
Benchmark scores via ARC Prize · pricing verified against LiteLLM + OpenRouter · refreshed daily · updated 15 July 2026

Every weekday

AI moves fast. Here's your debrief.

News, analysis, tools, and more.

For people who build with AI