ARC-AGI-2

Semi-private ARC-AGI-2 evaluation set.

Published by

ARC Prize

Category

Reasoning

Index weight

1.0

Models

187

Data as of

9 Sept 2026

ARC Prize Foundation, arcprize.org.

1. 1GPT-6 Astra95.0%
2. 2GPT-6 Astra93.3%
3. 3GPT-5.6 Sol92.5%
4. 4GPT-6 Astra92.1%
5. 5GPT-6 Astra92.1%
6. 6Claude Opus 590.4%
7. 7Claude Fable 5.190.0%
8. 8Claude Fable 5.190.0%
9. 9GPT-5.6 Sol90.0%
10. 10Claude Fable 589.2%
11. 11Claude Fable 5.188.8%
12. 12Claude Opus 588.3%
13. 13Claude Fable 588.3%
14. 14Claude Fable 587.5%
15. 15Claude Fable 5.186.3%

SearchAll providersAlibabaAnthropicDeepSeekGoogleMetaMiniMaxMistral AIMoonshot AIOpenAIThinking MachinesxAIZhipu AIOpen weights187 of 187

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	GPT-6 AstraOpenAI	95.0%	–	70.6	–
2	GPT-6 AstraOpenAI	93.3%	–	68.6	–
3	GPT-5.6 SolOpenAI	92.5%	–	65.5	–
4	GPT-6 AstraOpenAI	92.1%	–	69.5	–
5	GPT-6 AstraOpenAI	92.1%	–	67.4	–
6	Claude Opus 5Anthropic	90.4%	–	66.8	–
7	Claude Fable 5.1Anthropic	90.0%	–	70.3	–
8	Claude Fable 5.1Anthropic	90.0%	–	70.2	–
9	GPT-5.6 SolOpenAI	90.0%	–	67.5	–
10	Claude Fable 5Anthropic	89.2%	–	65.8	–
11	Claude Fable 5.1Anthropic	88.8%	–	69.7	–
12	Claude Opus 5Anthropic	88.3%	–	69.3	–
13	Claude Fable 5Anthropic	88.3%	–	–	–
14	Claude Fable 5Anthropic	87.5%	–	63.9	–
15	Claude Fable 5.1Anthropic	86.3%	–	68.0	–
16	GPT-5.6 SolOpenAI	85.4%	–	68.0	–
17	GPT-6 AstraOpenAI	85.4%	–	65.8	–
18	GPT-5.5OpenAI	85.0%	–	63.7	–
19	Gemini 3.7 FlashGoogle	84.6%	–	60.8	–
20	GPT-5.5 ProOpenAI	84.6%	–	–	–
21	Gemini 3 Deep ThinkGoogle	84.6%	–	–	–
22	GPT-5.5 ProOpenAI	84.2%	–	–	–
23	GPT-5.6 TerraOpenAI	83.9%	–	58.0	–
24	GPT-5.5OpenAI	83.3%	–	66.3	–
25	GPT-5.4 ProOpenAI	83.3%	–	61.1	–
26	Claude Fable 5Anthropic	82.5%	–	–	–
27	Claude Fable 5.1Anthropic	78.3%	–	65.9	–
28	Gemini 3.1 ProGoogle	77.1%	–	63.9	–
29	Claude Fable 5Anthropic	76.8%	–	–	–
30	Claude 4.7Anthropic	75.8%	–	–	–
31	GPT-5.6 TerraOpenAI	74.2%	–	64.2	–
32	GPT-5.4OpenAI	74.0%	–	61.3	–
33	GPT-5.2OpenAI	72.9%	–	61.5	–
34	Claude Opus 4.8Anthropic	72.1%	–	61.3	–
35	Gemini 3.5 FlashGoogle	72.1%	–	60.4	–
36	Claude Opus 4.8Anthropic	71.7%	–	–	–
37	GPT-5.5OpenAI	70.4%	–	64.9	–
38	Claude Opus 4.6Anthropic	69.2%	–	61.3	–
39	Claude Opus 4.6Anthropic	68.8%	–	54.1	–
40	Claude 4.7Anthropic	68.3%	–	–	–
41	GPT-5.4OpenAI	67.5%	–	58.8	–
42	Claude 4.7Anthropic	67.5%	–	–	–
43	GPT-5.6 SolOpenAI	67.1%	–	66.0	–
44	GPT-5.6 TerraOpenAI	67.1%	–	63.1	–
45	Grok 4.6xAI	67.1%	–	62.3	–
46	Claude Opus 4.6Anthropic	66.3%	–	–	–
47	Grok 4.6xAI	65.1%	–	58.7	–
48	Grok 4.20xAI	65.1%	–	54.3	–
49	Claude Opus 4.6Anthropic	64.6%	–	–	–
50	Gemini 3.7 FlashGoogle	63.8%	–	64.4	–
51	Claude Opus 4.8Anthropic	62.2%	–	–	–
52	Claude 4.7Anthropic	62.1%	–	–	–
53	DeepSeek V4 FlashDeepSeekopen	61.4%	max, 0731	54.7	–
54	Grok 4.6xAI	61.3%	–	63.8	–
55	DeepSeek V4 ProDeepSeekopen	61.3%	max, 0813	56.1	–
56	Kimi K3Moonshot AIopen	60.4%	–	62.7	–
57	Gemini 3.6 FlashGoogle	60.4%	–	58.6	–
58	Claude Sonnet 4.6Anthropic	60.4%	–	52.3	–
59	DeepSeek V4 ProDeepSeekopen	59.7%	high, 0813	60.0	–
60	GPT-6 AstraOpenAI	59.6%	–	64.4	–

Show 100 more