ARC-AGI-1

Semi-private ARC-AGI-1 evaluation set.

Published by

ARC Prize

Category

Reasoning

Index weight

0.5

Models

186

Data as of

9 Sept 2026

ARC Prize Foundation, arcprize.org.

1. 1GPT-6 Astra98.5%
2. 2GPT-6 Astra98.5%
3. 3Claude Fable 598.5%
4. 4Claude Fable 598.5%
5. 5Gemini 3.1 Pro98.0%
6. 6GPT-6 Astra97.5%
7. 7Claude Fable 5.197.5%
8. 8Claude Opus 597.5%
9. 9GPT-5.6 Sol97.5%
10. 10GPT-6 Astra97.5%
11. 11Claude Opus 597.5%
12. 12GPT-5.6 Sol97.0%
13. 13Claude Fable 5.196.5%
14. 14GPT-6 Astra96.5%
15. 15GPT-5.6 Sol96.5%

SearchAll providersAlibabaAnthropicDeepSeekGoogleMetaMiniMaxMistral AIMoonshot AIOpenAIThinking MachinesxAIZhipu AIOpen weights186 of 186

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	GPT-6 AstraOpenAI	98.5%	–	69.5	–
2	GPT-6 AstraOpenAI	98.5%	–	68.6	–
3	Claude Fable 5Anthropic	98.5%	–	65.8	–
4	Claude Fable 5Anthropic	98.5%	–	–	–
5	Gemini 3.1 ProGoogle	98.0%	–	63.9	–
6	GPT-6 AstraOpenAI	97.5%	–	70.6	–
7	Claude Fable 5.1Anthropic	97.5%	–	70.2	–
8	Claude Opus 5Anthropic	97.5%	–	69.3	–
9	GPT-5.6 SolOpenAI	97.5%	–	67.5	–
10	GPT-6 AstraOpenAI	97.5%	–	67.4	–
11	Claude Opus 5Anthropic	97.5%	–	66.8	–
12	GPT-5.6 SolOpenAI	97.0%	–	68.0	–
13	Claude Fable 5.1Anthropic	96.5%	–	70.3	–
14	GPT-6 AstraOpenAI	96.5%	–	65.8	–
15	GPT-5.6 SolOpenAI	96.5%	–	65.5	–
16	GPT-5.6 TerraOpenAI	96.5%	–	58.0	–
17	GPT-5.5 ProOpenAI	96.5%	–	–	–
18	Claude Fable 5.1Anthropic	96.0%	–	69.7	–
19	Gemini 3 Deep ThinkGoogle	96.0%	–	–	–
20	Claude Fable 5Anthropic	95.5%	–	63.9	–
21	Gemini 3.7 FlashGoogle	95.5%	–	60.8	–
22	GPT-5.5OpenAI	95.0%	–	63.7	–
23	GPT-5.5 ProOpenAI	95.0%	–	–	–
24	Claude Fable 5.1Anthropic	94.5%	–	68.0	–
25	GPT-5.5OpenAI	94.5%	–	66.3	–
26	Kimi K3Moonshot AIopen	94.5%	–	62.7	–
27	GPT-5.2OpenAI	94.5%	–	61.5	–
28	GPT-5.4 ProOpenAI	94.5%	–	61.1	–
29	GPT-5.6 TerraOpenAI	94.0%	–	64.2	–
30	Claude Opus 4.6Anthropic	94.0%	–	61.3	–
31	GPT-5.4OpenAI	93.7%	–	61.3	–
32	Claude 4.7Anthropic	93.5%	–	–	–
33	Claude Opus 4.6Anthropic	93.0%	–	54.1	–
34	GPT-5.4OpenAI	92.7%	–	58.8	–
35	GPT-5.6 SolOpenAI	92.5%	–	66.0	–
36	Gemini 3.5 FlashGoogle	92.5%	–	60.4	–
37	Claude Opus 4.8Anthropic	92.5%	–	59.4	–
38	Claude Fable 5Anthropic	92.5%	–	–	–
39	GPT-5.5OpenAI	92.2%	–	64.9	–
40	GPT-5.6 TerraOpenAI	92.0%	–	63.1	–
41	Claude Opus 4.8Anthropic	92.0%	–	61.3	–
42	Claude 4.7Anthropic	92.0%	–	–	–
43	Claude Opus 4.6Anthropic	92.0%	–	–	–
44	Claude Opus 4.8Anthropic	91.5%	–	–	–
45	Gemini 3.7 FlashGoogle	91.2%	–	64.4	–
46	Gemini 3.6 FlashGoogle	91.2%	–	58.6	–
47	Claude 4.7Anthropic	91.0%	–	–	–
48	Claude 4.7Anthropic	91.0%	–	–	–
49	GPT-5.6 LunaOpenAI	90.7%	–	56.5	–
50	Claude Fable 5Anthropic	90.5%	–	–	–
51	GPT-5.2 ProOpenAI	90.5%	–	–	–
52	DeepSeek V4 ProDeepSeekopen	90.5%	low, 0813	–	–
53	Claude Fable 5.1Anthropic	90.0%	–	65.9	–
54	DeepSeek V4 ProDeepSeekopen	90.0%	max, 0813	56.1	–
55	Grok 4.20xAI	89.5%	–	54.3	–
56	DeepSeek V4 FlashDeepSeekopen	89.0%	max, 0731	54.7	–
57	Claude Opus 4.8Anthropic	88.0%	–	–	–
58	GPT-5.6 LunaOpenAI	87.7%	–	59.3	–
59	Grok 4.6xAI	87.5%	–	63.8	–
60	Gemini 3 DeepGoogle	87.5%	–	–	–

Show 100 more