Abstraction and Reasoning Corpus for AGI v2 (ARC-AGI-2)

GPT-6 Astra (Max) leads the ARC-AGI-2 leaderboard on BenchLM's October 2026 update with 95%, ahead of GPT-6.1 Sol (Max) (94.2%) and Claude Opus 5.5 (High) (93.3%), across 290 configurations.

Data verified Captured 2026-10-25

ARC-AGI-2 tests whether a system can infer a transformation from example grids and apply it to a new grid. Compare accuracy with the evaluation split, attempt limit, and cost attached.

Successor track

ARC-AGI-3 evaluates interactive agentic reasoning rather than ARC-AGI-2's static grid puzzles. The two score tables are not directly comparable.

View ARC-AGI-3

Top models on ARC-AGI-2 — Captured 2026-10-25

As of Captured 2026-10-25, GPT-6 Astra (Max) leads the ARC-AGI-2 leaderboard with 95% , followed by GPT-6.1 Sol (Max) (94.2%) and Claude Opus 5.5 (High) (93.3%).

1Closed

GPT-6 Astra (Max)

OpenAI

Semi-Private evaluation · max reasoning

95%

Overall 84.43Context 1.05M

2Closed

GPT-6.1 Sol (Max)

OpenAI

Semi-Private evaluation · max reasoning

94.2%

Overall 84.38Context 1.05M

3Closed

Claude Opus 5.5 (High)

Anthropic

Semi-Private evaluation · high reasoning

93.3%

Overall 86.66Context 1M

290 configurationsReasoning25% of category scoreCurrentUpdated Captured 2026-10-25

Leaderboard (290 configurations)

ARC-AGI-2 leaderboard
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	GPT-6 Astra (Max)OpenAISemi-Private evaluation · max reasoning	95%	Not reported	Closed
2	GPT-6.1 Sol (Max)OpenAISemi-Private evaluation · max reasoning	94.2%	Not reported	Closed
3	Claude Opus 5.5 (High)AnthropicSemi-Private evaluation · high reasoning	93.3%	Not reported	Closed
4	GPT-6 Astra (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	93.3%	Not reported	Closed
5	Claude Opus 5.5 (XHigh)AnthropicSemi-Private evaluation · xhigh reasoning	92.5%	Not reported	Closed
6	GPT-5.6 Sol (Max)OpenAISemi-Private evaluation · max reasoning	92.5%	Not reported	Closed
7	GPT-6 Astra (High)OpenAISemi-Private evaluation · high reasoning	92.1%	Not reported	Closed
8	GPT-6 Astra (Medium)OpenAISemi-Private evaluation · medium reasoning	92.1%	Not reported	Closed
9	Claude Opus 5.5 (Max)AnthropicSemi-Private evaluation · max reasoning	91.7%	Not reported	Closed
10	GPT-6.1 Sol (High)OpenAISemi-Private evaluation · high reasoning	91.7%	Not reported	Closed
11	GPT-6.1 Sol (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	91.7%	Not reported	Closed
12	Gemini 4 Argon (High)GoogleSemi-Private evaluation · high reasoning	91.3%	Not reported	Closed
13	Claude Opus 5 (Max)AnthropicSemi-Private evaluation · max reasoning	90.4%	Not reported	Closed
14	Claude Fable 5.1 (Max)AnthropicSemi-Private evaluation · max reasoning	90%	Not reported	Unknown
15	Claude Fable 5.1 (XHigh)AnthropicSemi-Private evaluation · xhigh reasoning	90%	Not reported	Unknown
16	GPT-5.6 Sol (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	90%	Not reported	Closed
17	GPT-6 Sol (Max)OpenAISemi-Private evaluation · max reasoning	89.6%	Not reported	Closed
18	Claude Fable 5 (Max)AnthropicSemi-Private evaluation · max reasoning	89.2%	Not reported	Closed
19	Gemini 3.8 Flash (High)GoogleSemi-Private evaluation · high reasoning	89.2%	Not reported	Closed
20	Claude Fable 5.1 (High)AnthropicSemi-Private evaluation · high reasoning	88.8%	Not reported	Unknown
21	Claude Fable 5 (XHigh)AnthropicSemi-Private evaluation · xhigh reasoning	88.3%	Not reported	Closed
22	Claude Opus 5 (High)AnthropicSemi-Private evaluation · high reasoning	88.3%	Not reported	Closed
23	Claude Fable 5 (High)AnthropicSemi-Private evaluation · high reasoning	87.5%	Not reported	Closed
24	Claude Opus 5.5 (Medium)AnthropicSemi-Private evaluation · medium reasoning	87.5%	Not reported	Closed
25	Gemini 4 Argon (Medium)GoogleSemi-Private evaluation · medium reasoning	87.1%	Not reported	Closed
26	GPT-6.1 Sol (Medium)OpenAISemi-Private evaluation · medium reasoning	86.7%	Not reported	Closed
27	Claude Fable 5.1 (Medium)AnthropicSemi-Private evaluation · medium reasoning	86.3%	Not reported	Unknown
28	GPT-5.6 Sol (High)OpenAISemi-Private evaluation · high reasoning	85.4%	Not reported	Closed
29	GPT-6 Astra (Low)OpenAISemi-Private evaluation · low reasoning	85.4%	Not reported	Closed
30	GPT-5.5 (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	85%	Not reported	Unknown
31	Gemini 3.7 Flash (High)GoogleSemi-Private evaluation · high reasoning	84.6%	Not reported	Closed
32	Gemini 3 Deep Think (2/26)GoogleSemi-Private evaluation	84.6%	Not reported	Unknown
33	GPT-5.5 Pro (High)OpenAISemi-Private evaluation · high reasoning	84.6%	Not reported	Unknown
34	GPT-5.5 Pro (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	84.2%	Not reported	Unknown
35	GPT-5.6 Terra (Max)OpenAISemi-Private evaluation · max reasoning	83.9%	Not reported	Closed
36	Claude Sonnet 5.5 (XHigh)AnthropicSemi-Private evaluation · xhigh reasoning	83.9%	Not reported	Unknown
37	GPT-5.4 Pro (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	83.3%	Not reported	Unknown
38	GPT-5.5 (High)OpenAISemi-Private evaluation · high reasoning	83.3%	Not reported	Unknown
39	Gemini 3.8 Flash (Medium)GoogleSemi-Private evaluation · medium reasoning	82.9%	Not reported	Closed
40	Claude Fable 5 (Medium)AnthropicSemi-Private evaluation · medium reasoning	82.5%	Not reported	Closed
41	Claude Fable 5.1 (Low)AnthropicSemi-Private evaluation · low reasoning	78.3%	Not reported	Unknown
42	DeepSeek V4.1 Flash 1017 (Max)DeepSeekSemi-Private evaluation · max reasoning	78.1%	Not reported	Unknown
43	GPT-6 Sol (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	78.1%	Not reported	Closed
44	Gemini 3.8 Flash (Low)GoogleSemi-Private evaluation · low reasoning	77.5%	Not reported	Closed
45	Gemini 3.1 Pro (Preview)GoogleSemi-Private evaluation	77.1%	Not reported	Unknown
46	Claude Fable 5 (Low)AnthropicSemi-Private evaluation · low reasoning	76.8%	Not reported	Closed
47	Dots3-Note Preview (Max)Dots StudioSemi-Private evaluation · max reasoning	76.8%	Not reported	Unknown
48	GPT-6.1 Sol (Low)OpenAISemi-Private evaluation · low reasoning	76.7%	Not reported	Closed
49	Claude 4.7 (Max)AnthropicSemi-Private evaluation · max reasoning	75.8%	Not reported	Unknown
50	GPT-5.6 Terra (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	74.2%	Not reported	Closed
51	GPT-5.4 (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	74.0%	Not reported	Unknown
52	DeepSeek V4.1 Flash 1017 (Low)DeepSeekSemi-Private evaluation · low reasoning	73.6%	Not reported	Unknown
53	DeepSeek V4.1 Flash (Max)DeepSeekSemi-Private evaluation · max reasoning	72.9%	Not reported	Unknown
54	GPT-5.2 (Refine.)Johan LandSemi-Private evaluation	72.9%	Not reported	Unknown
55	Claude Opus 4.8 (High)AnthropicSemi-Private evaluation · high reasoning	72.1%	Not reported	Closed
56	Gemini 3.5 Flash (High)GoogleSemi-Private evaluation · high reasoning	72.1%	Not reported	Unknown
57	Claude Opus 4.8 (Medium)AnthropicSemi-Private evaluation · medium reasoning	71.7%	Not reported	Closed
58	GPT-5.5 (Medium)OpenAISemi-Private evaluation · medium reasoning	70.4%	Not reported	Unknown
59	Claude Opus 5.5 (Low)AnthropicSemi-Private evaluation · low reasoning	70.1%	Not reported	Closed
60	Claude Opus 4.6 (120K, High)AnthropicSemi-Private evaluation	69.2%	Not reported	Unknown
61	GPT-6 Sol (High)OpenAISemi-Private evaluation · high reasoning	68.9%	Not reported	Closed
62	Claude Opus 4.6 (120K, Max)AnthropicSemi-Private evaluation	68.8%	Not reported	Unknown
63	DeepSeek V4.1 Flash 1017 (High)DeepSeekSemi-Private evaluation · high reasoning	68.6%	Not reported	Unknown
64	Claude 4.7 (High)AnthropicSemi-Private evaluation · high reasoning	68.3%	Not reported	Unknown
65	Claude 4.7 (Medium)AnthropicSemi-Private evaluation · medium reasoning	67.5%	Not reported	Unknown
66	DeepSeek V4.1 Flash (High)DeepSeekSemi-Private evaluation · high reasoning	67.5%	Not reported	Unknown
67	GPT-5.4 (High)OpenAISemi-Private evaluation · high reasoning	67.5%	Not reported	Unknown
68	Claude Sonnet 5.5 (High)AnthropicSemi-Private evaluation · high reasoning	67.4%	Not reported	Unknown
69	Grok 4.6 (XHigh)xAISemi-Private evaluation · xhigh reasoning	67.1%	Not reported	Closed
70	GPT-5.6 Sol (Medium)OpenAISemi-Private evaluation · medium reasoning	67.1%	Not reported	Closed
71	GPT-5.6 Terra (High)OpenAISemi-Private evaluation · high reasoning	67.1%	Not reported	Closed
72	Claude Opus 4.6 (120K, Medium)AnthropicSemi-Private evaluation	66.3%	Not reported	Unknown
73	GLM-5.3-Flash (Max)Z.aiSemi-Private evaluation · max reasoning	65.8%	Not reported	Unknown
74	Grok 4.20 (Reasoning)xAISemi-Private evaluation	65.1%	Not reported	Unknown
75	Grok 4.6 (High)xAISemi-Private evaluation · high reasoning	65.1%	Not reported	Closed
76	Claude Opus 4.6 (120K, Low)AnthropicSemi-Private evaluation	64.6%	Not reported	Unknown
77	Gemini 3.7 Flash (Medium)GoogleSemi-Private evaluation · medium reasoning	63.8%	Not reported	Closed
78	Claude Opus 4.8 (Low)AnthropicSemi-Private evaluation · low reasoning	62.2%	Not reported	Closed
79	Claude 4.7 (Low)AnthropicSemi-Private evaluation · low reasoning	62.1%	Not reported	Unknown
80	DeepSeek V4 Flash 0731 (Max)DeepSeekSemi-Private evaluation · max reasoning	61.4%	Not reported	Open
81	Grok 4.7 (XHigh)xAISemi-Private evaluation · xhigh reasoning	61.4%	Not reported	Closed
82	Claude Haiku 5.5 (XHigh)AnthropicSemi-Private evaluation · xhigh reasoning	61.3%	Not reported	Closed
83	DeepSeek V4 Pro 0813 (Max)DeepSeekSemi-Private evaluation · max reasoning	61.3%	Not reported	Open
84	Grok 4.6 (Medium)xAISemi-Private evaluation · medium reasoning	61.3%	Not reported	Closed
85	DeepSeek V4.1 Flash (Low)DeepSeekSemi-Private evaluation · low reasoning	60.6%	Not reported	Unknown
86	Claude Sonnet 4.6 (High)AnthropicSemi-Private evaluation · high reasoning	60.4%	Not reported	Unknown
87	Gemini 3.6 Flash (High)GoogleSemi-Private evaluation · high reasoning	60.4%	Not reported	Closed
88	Kimi K3 (Max)Moonshot AISemi-Private evaluation · max reasoning	60.4%	Not reported	Pending
89	DeepSeek V4 Pro 0813 (High)DeepSeekSemi-Private evaluation · high reasoning	59.7%	Not reported	Open
90	GPT-5.6 Luna 2026-07-30 (Max)OpenAISemi-Private evaluation · max reasoning	59.6%	Not reported	Unknown
91	GPT-6 Astra (None)OpenAISemi-Private evaluation · none reasoning	59.6%	Not reported	Closed
92	GPT-5.6 Luna (Max)OpenAISemi-Private evaluation · max reasoning	59.5%	Not reported	Closed
93	GPT-6 Luna (Max)OpenAISemi-Private evaluation · max reasoning	59.3%	Not reported	Closed
94	Grok 4.7 (Medium)xAISemi-Private evaluation · medium reasoning	58.8%	Not reported	Closed
95	Grok 4.7 (High)xAISemi-Private evaluation · high reasoning	58.3%	Not reported	Closed
96	Claude Sonnet 4.6 (Max)AnthropicSemi-Private evaluation · max reasoning	58.3%	Not reported	Unknown
97	GPT-6 Sol (Medium)OpenAISemi-Private evaluation · medium reasoning	57.8%	Not reported	Closed
98	Muse Spark Open (XHigh)Meta AISemi-Private evaluation · xhigh reasoning	57.5%	Not reported	Unknown
99	DeepSeek V4 Pro 0813 (Low)DeepSeekSemi-Private evaluation · low reasoning	56.3%	Not reported	Open
100	DeepSeek V4 Flash 0731 (High)DeepSeekSemi-Private evaluation · high reasoning	56.0%	Not reported	Open
101	GPT-5.4 (Medium)OpenAISemi-Private evaluation · medium reasoning	55.4%	Not reported	Unknown
102	Kimi K3 (High)Moonshot AISemi-Private evaluation · high reasoning	55%	Not reported	Pending
103	GPT-5.2 Pro (High)OpenAISemi-Private evaluation · high reasoning	54.2%	Not reported	Unknown
104	Gemini 3 Pro (Refine.)PoetiqSemi-Private evaluation	54%	Not reported	Unknown
105	Gemini 3.7 Flash (Low)GoogleSemi-Private evaluation · low reasoning	52.9%	Not reported	Closed
106	GPT-5.2 (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	52.9%	Not reported	Unknown
107	Grok 4.5 (High)xAISemi-Private evaluation · high reasoning	52.6%	Not reported	Closed
108	Grok 4.5 (Medium)xAISemi-Private evaluation · medium reasoning	52.6%	Not reported	Closed
109	Gemini 3.6 Flash (Medium)GoogleSemi-Private evaluation · medium reasoning	50.4%	Not reported	Closed
110	GLM-5.3-Flash (High)Z.aiSemi-Private evaluation · high reasoning	50.1%	Not reported	Unknown
111	GPT-5.6 Luna (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	47.6%	Not reported	Closed
112	Muse Spark Open (High)Meta AISemi-Private evaluation · high reasoning	47.5%	Not reported	Unknown
113	DeepSeek V4 Flash 0731 (Low)DeepSeekSemi-Private evaluation · low reasoning	46.0%	Not reported	Open
114	Gemini 3 Deep Think (Preview) ²GoogleSemi-Private evaluation	45.1%	Not reported	Unknown
115	GPT-5.6 Luna 2026-07-30 (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	43.5%	Not reported	Unknown
116	GPT-5.2 (High)OpenAISemi-Private evaluation · high reasoning	43.3%	Not reported	Unknown
117	GPT-5.6 Sol (Low)OpenAISemi-Private evaluation · low reasoning	42.5%	Not reported	Closed
118	Qwen3.8-27B (XHigh)AlibabaSemi-Private evaluation · xhigh reasoning	42.4%	Not reported	Unknown
119	Gemini 4 Argon (Low)GoogleSemi-Private evaluation · low reasoning	42.1%	Not reported	Closed
120	GPT-6 Luna (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	41.9%	Not reported	Closed
121	Inkling Small (XHigh)Thinking Machines LabSemi-Private evaluation · xhigh reasoning	40.1%	Not reported	Open
122	GPT-5.2 Pro (Medium)OpenAISemi-Private evaluation · medium reasoning	38.5%	Not reported	Unknown
123	Opus 4.5 (Thinking, 64K)AnthropicSemi-Private evaluation	37.6%	Not reported	Unknown
124	GPT-5.6 Terra (Medium)OpenAISemi-Private evaluation · medium reasoning	37.5%	Not reported	Closed
125	InklingThinking MachinesSemi-Private evaluation	36.5%	Not reported	Unknown
126	Gemini 3 Flash Preview (High)GoogleSemi-Private evaluation · high reasoning	33.6%	Not reported	Unknown
127	GPT-5.6 Luna 2026-07-30 (High)OpenAISemi-Private evaluation · high reasoning	33.3%	Not reported	Unknown
128	GPT-5.5 (Low)OpenAISemi-Private evaluation · low reasoning	33.3%	Not reported	Unknown
129	Grok 4.5 (Low)xAISemi-Private evaluation · low reasoning	33.1%	Not reported	Closed
130	Inkling Small (High)Thinking Machines LabSemi-Private evaluation · high reasoning	33.1%	Not reported	Open
131	Claude Sonnet 5.5 (Medium)AnthropicSemi-Private evaluation · medium reasoning	32.8%	Not reported	Unknown
132	GPT-6 Sol (Low)OpenAISemi-Private evaluation · low reasoning	31.5%	Not reported	Closed
133	GPT-6 Luna (High)OpenAISemi-Private evaluation · high reasoning	31.4%	Not reported	Closed
134	Gemini 3 ProGoogleSemi-Private evaluation	31.1%	Not reported	Unknown
135	Claude Haiku 5.5 (High)AnthropicSemi-Private evaluation · high reasoning	30.7%	Not reported	Closed
136	Gemini 3.6 Flash (Low)GoogleSemi-Private evaluation · low reasoning	30.4%	Not reported	Closed
137	Grok 4 (Refine.)J. BermanSemi-Private evaluation	29.4%	Not reported	Unknown
138	GPT-5.6 Luna (High)OpenAISemi-Private evaluation · high reasoning	29.3%	Not reported	Closed
139	GPT-5.4 (Low)OpenAISemi-Private evaluation · low reasoning	29.2%	Not reported	Unknown
140	GLM-5.3-Flash (Low)Z.aiSemi-Private evaluation · low reasoning	27.9%	Not reported	Unknown
141	NVARCARC Prize 2025Semi-Private evaluation	27.6%	Not reported	Unknown
142	Grok 4.6 (Low)xAISemi-Private evaluation · low reasoning	27.6%	Not reported	Closed
143	Muse Spark Open (Medium)Meta AISemi-Private evaluation · medium reasoning	27.5%	Not reported	Unknown
144	GPT-5.2 (Medium)OpenAISemi-Private evaluation · medium reasoning	26.7%	Not reported	Unknown
145	Claude Sonnet 5.5 (Low)AnthropicSemi-Private evaluation · low reasoning	24.0%	Not reported	Unknown
146	Beam 1.0 (XHigh)ReflectionSemi-Private evaluation · xhigh reasoning	23.5%	Not reported	Unknown
147	Opus 4.5 (Thinking, 16K)AnthropicSemi-Private evaluation	22.8%	Not reported	Unknown
148	Qwen3.8-27B (Low)AlibabaSemi-Private evaluation · low reasoning	22.8%	Not reported	Unknown
149	GLM-5.2Z.aiSemi-Private evaluation	22.8%	Not reported	Unknown
150	Beam 1.0 (High)ReflectionSemi-Private evaluation · high reasoning	20%	Not reported	Unknown
151	GPT-5.4 Mini (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	18.9%	Not reported	Unknown
152	GPT-5.6 Terra (Low)OpenAISemi-Private evaluation · low reasoning	18.8%	Not reported	Closed
153	GPT-5 ProOpenAISemi-Private evaluation	18.3%	Not reported	Unknown
154	GPT-6 Luna (Medium)OpenAISemi-Private evaluation · medium reasoning	18.1%	Not reported	Closed
155	GPT-5.1 (Thinking, High)OpenAISemi-Private evaluation	17.6%	Not reported	Unknown
156	Grok 4.7 (Low)xAISemi-Private evaluation · low reasoning	16.7%	Not reported	Closed
157	Grok 4 (Thinking)xAISemi-Private evaluation	16.0%	Not reported	Unknown
158	Opus 4.5 (Thinking, 8K)AnthropicSemi-Private evaluation	13.9%	Not reported	Unknown
159	Inkling Small (Medium)Thinking Machines LabSemi-Private evaluation · medium reasoning	13.6%	Not reported	Open
160	Claude Sonnet 4.5 (Thinking 32K)AnthropicSemi-Private evaluation	13.6%	Not reported	Unknown
161	Qwen3.8-27B (Medium)AlibabaSemi-Private evaluation · medium reasoning	13.2%	Not reported	Unknown
162	GPT-5.4 Mini (High)OpenAISemi-Private evaluation · high reasoning	13.2%	Not reported	Unknown
163	Gemini 3 Flash Preview (Medium)GoogleSemi-Private evaluation · medium reasoning	12.8%	Not reported	Unknown
164	Kimi K3 (Low)Moonshot AISemi-Private evaluation · low reasoning	12.4%	Not reported	Pending
165	Muse Spark Open (Low)Meta AISemi-Private evaluation · low reasoning	12.2%	Not reported	Unknown
166	Kimi K2.5Moonshot AISemi-Private evaluation	11.8%	Not reported	Unknown
167	Gemini 3.5 Flash-Lite (High)GoogleSemi-Private evaluation · high reasoning	10.3%	Not reported	Unknown
168	GPT-5 (High)OpenAISemi-Private evaluation · high reasoning	9.9%	Not reported	Unknown
169	GPT-5.2 (Low)OpenAISemi-Private evaluation · low reasoning	9.7%	Not reported	Unknown
170	Claude Haiku 5.5 (Medium)AnthropicSemi-Private evaluation · medium reasoning	9.4%	Not reported	Closed
171	Gemini 3.5 Flash (Minimal)GoogleSemi-Private evaluation · minimal reasoning	8.9%	Not reported	Unknown
172	Claude Opus 4 (Thinking 16K)AnthropicSemi-Private evaluation	8.6%	Not reported	Unknown
173	Opus 4.5 (Thinking, None)AnthropicSemi-Private evaluation	7.8%	Not reported	Unknown
174	GPT-5 (Medium)OpenAISemi-Private evaluation · medium reasoning	7.5%	Not reported	Unknown
175	GPT-5.6 Luna 2026-07-30 (Medium)OpenAISemi-Private evaluation · medium reasoning	7.4%	Not reported	Unknown
176	GPT-5.6 Luna (Medium)OpenAISemi-Private evaluation · medium reasoning	7.4%	Not reported	Closed
177	Claude Sonnet 4.5 (Thinking 16K)AnthropicSemi-Private evaluation	6.9%	Not reported	Unknown
178	Claude Sonnet 4.5 (Thinking 8K)AnthropicSemi-Private evaluation	6.9%	Not reported	Unknown
179	GPT-5.1 (Thinking, Medium)OpenAISemi-Private evaluation	6.5%	Not reported	Unknown
180	o3 (High)OpenAISemi-Private evaluation · high reasoning	6.5%	Not reported	Unknown
181	Beam 1.0 (Medium)ReflectionSemi-Private evaluation · medium reasoning	6.4%	Not reported	Unknown
182	Tiny Recursion Model (TRM)BespokeSemi-Private evaluation	6.3%	Not reported	Unknown
183	o4-mini (High)OpenAISemi-Private evaluation · high reasoning	6.1%	Not reported	Unknown
184	Claude Sonnet 4 (Thinking 16K)AnthropicSemi-Private evaluation	5.9%	Not reported	Unknown
185	Claude Sonnet 4.5 (Thinking 1K)AnthropicSemi-Private evaluation	5.8%	Not reported	Unknown
186	GPT-5.4 Nano (XHigh)OpenAISemi-Private evaluation · xhigh reasoning	5.7%	Not reported	Unknown
187	Grok 4 (Fast Reasoning)xAISemi-Private evaluation	5.3%	Not reported	Unknown
188	Gemini 3.5 Flash-Lite (Medium)GoogleSemi-Private evaluation · medium reasoning	5.3%	Not reported	Unknown
189	GPT-5.6 Luna (Low)OpenAISemi-Private evaluation · low reasoning	5.1%	Not reported	Closed
190	Inkling Small (Low)Thinking Machines LabSemi-Private evaluation · low reasoning	4.9%	Not reported	Open
191	Gemini 2.5 Pro (Thinking 32K)GoogleSemi-Private evaluation	4.9%	Not reported	Unknown
192	GLM-5Z.aiSemi-Private evaluation	4.9%	Not reported	Unknown
193	Minimax M2.5MinimaxSemi-Private evaluation	4.9%	Not reported	Unknown
194	o3-Pro (High)OpenAISemi-Private evaluation · high reasoning	4.9%	Not reported	Unknown
195	GPT-6 Luna (Low)OpenAISemi-Private evaluation · low reasoning	4.6%	Not reported	Closed
196	Claude Opus 4 (Thinking 8K)AnthropicSemi-Private evaluation	4.5%	Not reported	Unknown
197	GPT-5 Mini (High)OpenAISemi-Private evaluation · high reasoning	4.4%	Not reported	Unknown
198	GPT-5.4 Mini (Medium)OpenAISemi-Private evaluation · medium reasoning	4.4%	Not reported	Unknown
199	Claude Haiku 4.5 (Thinking 32K)AnthropicSemi-Private evaluation	4.0%	Not reported	Unknown
200	Deepseek V3.2DeepSeekSemi-Private evaluation	4.0%	Not reported	Unknown
201	Gemini 2.5 Pro (Thinking 16K)GoogleSemi-Private evaluation	4.0%	Not reported	Unknown
202	GPT-5 Mini (Medium)OpenAISemi-Private evaluation · medium reasoning	4.0%	Not reported	Unknown
203	Inkling Small (Minimal)Thinking Machines LabSemi-Private evaluation · minimal reasoning	4.0%	Not reported	Open
204	Claude Sonnet 4.5AnthropicSemi-Private evaluation	3.8%	Not reported	Unknown
205	GPT-5.6 Luna 2026-07-30 (Low)OpenAISemi-Private evaluation · low reasoning	3.6%	Not reported	Unknown
206	GPT-5.4 Nano (High)OpenAISemi-Private evaluation · high reasoning	3.6%	Not reported	Unknown
207	Gemini 3 Flash Preview (Minimal)GoogleSemi-Private evaluation · minimal reasoning	3.3%	Not reported	Unknown
208	o3-mini (High)OpenAISemi-Private evaluation · high reasoning	3.0%	Not reported	Unknown
209	o3 (Medium)OpenAISemi-Private evaluation · medium reasoning	3.0%	Not reported	Unknown
210	Gemini 2.5 Pro (Thinking 8K)GoogleSemi-Private evaluation	2.9%	Not reported	Unknown
211	Claude Haiku 4.5 (Thinking 16K)AnthropicSemi-Private evaluation	2.8%	Not reported	Unknown
212	Beam 1.0 (Low)ReflectionSemi-Private evaluation · low reasoning	2.8%	Not reported	Unknown
213	Claude Haiku 5.5 (Low)AnthropicSemi-Private evaluation · low reasoning	2.8%	Not reported	Closed
214	Gemini 3.6 Flash (Minimal)GoogleSemi-Private evaluation · minimal reasoning	2.6%	Not reported	Closed
215	GPT-5 Nano (High)OpenAISemi-Private evaluation · high reasoning	2.6%	Not reported	Unknown
216	Gemini 2.5 Flash (Preview) (Thinking 24K)GoogleSemi-Private evaluation	2.5%	Not reported	Unknown
217	ARChitectsARC Prize 2024Semi-Private evaluation	2.5%	Not reported	Unknown
218	o4-mini (Medium)OpenAISemi-Private evaluation · medium reasoning	2.4%	Not reported	Unknown
219	Gemini 2.5 Flash (Preview) (Thinking 1K)GoogleSemi-Private evaluation	2.2%	Not reported	Unknown
220	Claude Sonnet 4 (Thinking 8K)AnthropicSemi-Private evaluation	2.1%	Not reported	Unknown
221	Gemini 2.5 Flash (Preview) (Thinking 8K)GoogleSemi-Private evaluation	2.1%	Not reported	Unknown
222	DeepSeek V4 Flash 0731 (None)DeepSeekSemi-Private evaluation · none reasoning	2.1%	Not reported	Open
223	DeepSeek V4.1 Flash 1017 (None)DeepSeekSemi-Private evaluation · none reasoning	2.1%	Not reported	Unknown
224	o3-mini (Medium)OpenAISemi-Private evaluation · medium reasoning	2.1%	Not reported	Unknown
225	o3-Pro (Low)OpenAISemi-Private evaluation · low reasoning	2.0%	Not reported	Unknown
226	o3 (Low)OpenAISemi-Private evaluation · low reasoning	2.0%	Not reported	Unknown
227	Gemini 2.5 Flash (Preview) (Thinking 16K)GoogleSemi-Private evaluation	2.0%	Not reported	Unknown
228	Nemotron 3.5 Super VL (Medium)NVIDIASemi-Private evaluation · medium reasoning	1.9%	Not reported	Unknown
229	GPT-5 (Low)OpenAISemi-Private evaluation · low reasoning	1.9%	Not reported	Unknown
230	GPT-5.1 (Thinking, Low)OpenAISemi-Private evaluation	1.9%	Not reported	Unknown
231	GPT-5.4 Nano (Medium)OpenAISemi-Private evaluation · medium reasoning	1.9%	Not reported	Unknown
232	o3-Pro (Medium)OpenAISemi-Private evaluation · medium reasoning	1.9%	Not reported	Unknown
233	Gemini 2.5 Flash (Preview)GoogleSemi-Private evaluation	1.7%	Not reported	Unknown
234	Claude Haiku 4.5 (Thinking 8K)AnthropicSemi-Private evaluation	1.7%	Not reported	Unknown
235	GPT-5 Mini (Minimal)OpenAISemi-Private evaluation · minimal reasoning	1.7%	Not reported	Unknown
236	o4-mini (Low)OpenAISemi-Private evaluation · low reasoning	1.7%	Not reported	Unknown
237	GPT-6 Sol (None)OpenAISemi-Private evaluation · none reasoning	1.7%	Not reported	Closed
238	IcecuberARC Prize 2024Semi-Private evaluation	1.6%	Not reported	Unknown
239	GPT-5.4 Nano (Low)OpenAISemi-Private evaluation · low reasoning	1.5%	Not reported	Unknown
240	Gemini 3.5 Flash-Lite (Low)GoogleSemi-Private evaluation · low reasoning	1.5%	Not reported	Unknown
241	Nemotron 3.5 Super VL (High)NVIDIASemi-Private evaluation · high reasoning	1.5%	Not reported	Unknown
242	Qwen3.8-27B (None)AlibabaSemi-Private evaluation · none reasoning	1.5%	Not reported	Unknown
243	Deepseek R1DeepSeekSemi-Private evaluation	1.3%	Not reported	Unknown
244	Gemini 2.0 FlashGoogleSemi-Private evaluation	1.3%	Not reported	Unknown
245	Claude Opus 4AnthropicSemi-Private evaluation	1.3%	Not reported	Unknown
246	Claude Sonnet 4AnthropicSemi-Private evaluation	1.3%	Not reported	Unknown
247	Codex Mini (Latest)OpenAISemi-Private evaluation	1.3%	Not reported	Unknown
248	Claude Haiku 4.5AnthropicSemi-Private evaluation	1.3%	Not reported	Unknown
249	Claude Haiku 4.5 (Thinking 1K)AnthropicSemi-Private evaluation	1.3%	Not reported	Unknown
250	DeepSeek V4.1 Flash (None)DeepSeekSemi-Private evaluation · none reasoning	1.3%	Not reported	Unknown
251	Gemini 3 Flash Preview (Low)GoogleSemi-Private evaluation · low reasoning	1.3%	Not reported	Unknown
252	Qwen3-235b-a22b Instruct (25/07)AlibabaSemi-Private evaluation	1.3%	Not reported	Unknown
253	Deepseek R1 (05/28)DeepSeekSemi-Private evaluation	1.1%	Not reported	Unknown
254	GPT-5.4 Mini (Low)OpenAISemi-Private evaluation · low reasoning	1.1%	Not reported	Unknown
255	Claude 3.7 (8K)AnthropicSemi-Private evaluation	0.9%	Not reported	Unknown
256	GPT-5 Nano (Medium)OpenAISemi-Private evaluation · medium reasoning	0.9%	Not reported	Unknown
257	Claude Sonnet 4 (Thinking 1K)AnthropicSemi-Private evaluation	0.8%	Not reported	Unknown
258	DeepSeek V4 Pro 0813 (None)DeepSeekSemi-Private evaluation · none reasoning	0.8%	Not reported	Open
259	Gemini 3.5 Flash-Lite (Minimal)GoogleSemi-Private evaluation · minimal reasoning	0.8%	Not reported	Unknown
260	GPT-5.6 Luna 2026-07-30 (None)OpenAISemi-Private evaluation · none reasoning	0.8%	Not reported	Unknown
261	Nemotron 3.5 Super VL (None)NVIDIASemi-Private evaluation · none reasoning	0.8%	Not reported	Unknown
262	GPT-5 Mini (Low)OpenAISemi-Private evaluation · low reasoning	0.8%	Not reported	Unknown
263	GPT-5.2OpenAISemi-Private evaluation	0.8%	Not reported	Unknown
264	o1-miniOpenAISemi-Private evaluation	0.8%	Not reported	Unknown
265	Gemini 1.5 ProGoogleSemi-Private evaluation	0.8%	Not reported	Unknown
266	GPT-4.5OpenAISemi-Private evaluation	0.8%	Not reported	Unknown
267	Claude 3.7 (16K)AnthropicSemi-Private evaluation	0.7%	Not reported	Unknown
268	GPT-4.1OpenAISemi-Private evaluation	0.4%	Not reported	Unknown
269	GPT-5.1 (Thinking, None)OpenAISemi-Private evaluation	0.4%	Not reported	Unknown
270	Grok 3 Mini (Low)xAISemi-Private evaluation · low reasoning	0.4%	Not reported	Unknown
271	Claude 3.7 (1K)AnthropicSemi-Private evaluation	0.4%	Not reported	Unknown
272	Claude 3.7AnthropicSemi-Private evaluation	0%	Not reported	Unknown
273	Claude Opus 4 (Thinking 1K)AnthropicSemi-Private evaluation	0%	Not reported	Unknown
274	Gemini 2.5 Pro (Thinking 1K)GoogleSemi-Private evaluation	0%	Not reported	Unknown
275	GPT-4.1-MiniOpenAISemi-Private evaluation	0%	Not reported	Unknown
276	GPT-4.1-NanoOpenAISemi-Private evaluation	0%	Not reported	Unknown
277	GPT-4oOpenAISemi-Private evaluation	0%	Not reported	Unknown
278	GPT-4o-miniOpenAISemi-Private evaluation	0%	Not reported	Unknown
279	GPT-5 (Minimal)OpenAISemi-Private evaluation · minimal reasoning	0%	Not reported	Unknown
280	GPT-5 Nano (Low)OpenAISemi-Private evaluation · low reasoning	0%	Not reported	Unknown
281	GPT-5 Nano (Minimal)OpenAISemi-Private evaluation · minimal reasoning	0%	Not reported	Unknown
282	GPT-6 Luna (None)OpenAISemi-Private evaluation · none reasoning	0%	Not reported	Closed
283	Grok 3xAISemi-Private evaluation	0%	Not reported	Unknown
284	Inkling Small (None)Thinking Machines LabSemi-Private evaluation · none reasoning	0%	Not reported	Open
285	Llama 4 MaverickMetaSemi-Private evaluation	0%	Not reported	Unknown
286	Llama 4 ScoutMetaSemi-Private evaluation	0%	Not reported	Unknown
287	Magistral MediumMistralSemi-Private evaluation	0%	Not reported	Unknown
288	Magistral Medium (Thinking)MistralSemi-Private evaluation	0%	Not reported	Unknown
289	Magistral SmallMistralSemi-Private evaluation	0%	Not reported	Unknown
290	o3-mini (Low)OpenAISemi-Private evaluation · low reasoning	0%	Not reported	Unknown

How to read this leaderboard

ARC-AGI-2 is a static grid-puzzle reasoning benchmark released in 2025. Its public, semi-private, and private evaluation sets each contain 120 tasks; the training set contains 1,000 tasks.

Variant check: ARC-AGI-1, ARC-AGI-2, and the interactive ARC-AGI-3 track are different tests. Match the evaluation split and attempt rule within ARC-AGI-2 before treating a score gap as a model difference.

Use ARC-AGI-2 to examine adaptation to unfamiliar grid rules. The benchmark’s efficiency reporting matters: two systems can reach similar accuracy with different amounts of inference compute. Catalog token prices do not reproduce the reported cost per puzzle.

Operator receipt: 290 sourced rows are currently displayable on this page; the leading published row is GPT-6 Astra (Max) at 95%.

Honest limit: A result belongs to the evaluated system, including its adaptation or search procedure. It is not a general measure of every reasoning task, and it does not establish performance on the interactive ARC-AGI-3 track.

ARC-AGI-3 results Reasoning benchmark tables Benchmark confidence

ARC Prize ARC-AGI-2 task sets and efficiency rules

Semi-Private configurations

These are ARC Prize's published Semi-Private results. Every published reasoning-effort configuration stays separate; the table does not combine efforts or evaluation splits.

ARC-AGI-1 and ARC-AGI-2 use different task sets and remain separate from the interactive ARC-AGI-3 track.

Model pages use the highest declared published reasoning effort within each harness, rather than selecting whichever effort scored highest. Hidden or unpublished incomplete runs are not assigned a score.

Snapshot

ARC Prize leaderboard Original aggregate results

According to BenchLM.ai, GPT-6 Astra (Max) leads the ARC-AGI-2 benchmark with a score of 95%, followed by GPT-6.1 Sol (Max) (94.2%) and Claude Opus 5.5 (High) (93.3%). The top models are clustered within 1.7 points, suggesting this benchmark is nearing saturation for frontier models.

290 configurations have been evaluated on ARC-AGI-2. The benchmark falls in the Reasoning category. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking. The Reasoning leaderboard ranks models by a weighted category score, and ARC-AGI-2 contributes 25% of it. BenchAlign v5.8 also uses it in the overall ranking.

About ARC-AGI-2

Year

2025

Tasks

Visual pattern completion and abstract reasoning

Format

Grid transformation puzzles with novel rules

Difficulty

Expert-level — hardest public reasoning benchmark

ARC-AGI-2 extends the original ARC benchmark with harder puzzles designed to test fluid intelligence. Average individual human performance is 66%, the human panel completion rate is 100%, and the grand prize threshold is greater than 85%. Model results should be read with their exact evaluation set and reasoning effort.

ARC-AGI-2: A Harder General Intelligence Benchmark Public benchmark source

Freshness and provenance

Version

ARC-AGI 2

Refresh cadence

Static

Staleness state

Current

Question availability

Public benchmark set

Current

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does ARC-AGI-2 measure?

ARC-AGI-2 tests whether a system can infer a transformation from example grids and apply it to a new grid. Compare accuracy with the evaluation split, attempt limit, and cost attached.

Which model leads the published ARC-AGI-2 snapshot?

GPT-6 Astra (Max) currently leads the published ARC-AGI-2 snapshot with 95% semi-private score. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking.

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

The Captured 2026-10-25 snapshot contains 290 AI models.

Learn More

Read our explainer: ARC-AGI-2 benchmark deep dive

Last updated: Captured 2026-10-25 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 6,100+ subscribers.

One email each week. Unsubscribe anytime.