Knowledge & reasoning

ARC-AGI-2 leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on ARC-AGI-2 is 95.0% by GPT-6 Astra. 200 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	95.0%	measured
2	GPT-6.1 Sol	94.2%	measured
3	GPT-5.6 Sol	92.5%	measured
4	Claude Opus 5.5	91.7%	measured
5	Sakana Fugu	91.5%	estimated ± 10.1 pp, low confidence
6	Sakana Fugu-Ultra	91.5%	estimated ± 10.1 pp, low confidence
7	Claude Opus 5	90.4%	measured
8	Claude Fable 5.1	90.0%	measured
9	GPT-6 Sol	89.6%	measured
10	Claude Fable 5	89.2%	measured
11	Gemini 3.8 Flash	89.2%	measured
12	GPT-5.5	85.0%	measured
13	Claude Sonnet 5.5	84.8%	estimated ± 13.0 pp, low confidence
14	Gemini 3.7 Flash	84.6%	measured
15	GPT-5.5 Pro	84.2%	measured
16	GPT-5.6 Terra	83.9%	measured
17	GPT-5.4 Pro	83.3%	measured
18	Gemini 4 Argon	83.0%	estimated ± 14.2 pp, low confidence
19	dots3-note Preview	81.4%	measured
20	Pareto 26.9	80.9%	estimated ± 13.0 pp, low confidence
21	Muse Spark 1.3	80.9%	estimated ± 14.2 pp, low confidence
22	Grok 4.7	79.8%	estimated ± 14.2 pp, low confidence
23	MiMo-V2.6-Pro	79.7%	estimated ± 14.2 pp, low confidence
24	Claude Mythos 5	79.5%	estimated ± 11.6 pp, low confidence
25	Qwen3.8 Max Preview	79.1%	estimated ± 14.2 pp, low confidence
26	Claude Haiku 5.5	78.0%	estimated ± 13.0 pp, low confidence
27	Gemini 3.1 Pro	77.1%	measured
28	Claude Opus 4.7 (Adaptive)	75.8%	measured
29	Ling 3.1 Flash	74.9%	estimated ± 14.2 pp, low confidence
30	Step 5 Preview	74.7%	estimated ± 10.1 pp, low confidence
31	GPT-5.4	74.0%	measured
32	MiniMax M3	72.7%	estimated ± 12.6 pp, low confidence
33	Gemini 3.5 Flash	72.1%	measured
34	Claude Opus 4.8	72.1%	measured
35	Mistral Large 4	71.1%	estimated ± 14.2 pp, low confidence
36	MiMo-V2.6-Flash	70.3%	estimated ± 14.2 pp, low confidence
37	Ornith-1.5-397B	70.1%	estimated ± 10.1 pp, low confidence
38	Qwen3.8 Max	68.9%	estimated ± 10.1 pp, low confidence
39	Claude Opus 4.6 (Adaptive)	68.8%	measured
40	Pareto 26.10 Preview	67.7%	estimated ± 10.1 pp, low confidence
41	Qwen3.7 Max	67.7%	estimated ± 10.1 pp, low confidence
42	Grok 4.6	67.1%	measured
43	Hy4 preview	67.1%	estimated ± 10.1 pp, low confidence
44	Muse Spark 1.2	66.8%	estimated ± 13.0 pp, low confidence
45	Muse Spark 1.1	66.2%	estimated ± 12.6 pp, low confidence
46	Qwen3.8-Flash-Next	63.8%	estimated ± 10.1 pp, low confidence
47	Claude Sonnet 5	62.5%	estimated ± 3.1 pp, low confidence
48	Claude Opus 4.7	61.8%	estimated ± 12.6 pp, low confidence
49	DeepSeek V4 Flash 0731	61.4%	measured
50	DeepSeek V4 Pro 0813	61.3%	measured
51	GLM-5.2	61.2%	estimated ± 10.1 pp, low confidence
52	Gemini 3.6 Flash	60.4%	measured
53	Kimi K3	60.4%	measured
54	Qwen3.8-Omni-Flash	60.3%	estimated ± 10.1 pp, low confidence
55	DeepSeek V4.1 Flash	59.8%	estimated ± 10.1 pp, low confidence
56	GPT-5.6 Luna	59.5%	measured
57	GPT-6 Luna	59.3%	measured
58	GPT-5.3 Codex	58.9%	estimated ± 14.2 pp, low confidence
59	Claude Sonnet 4.6	58.3%	measured
60	Kimi K2.6	57.9%	estimated ± 10.1 pp, low confidence
61	Beam	57.9%	estimated ± 10.1 pp, low confidence
62	Qwen3.7 Plus	57.0%	estimated ± 10.1 pp, low confidence
63	Qwen3.6 Plus	55.6%	estimated ± 11.6 pp, low confidence
64	Interfaze Beta	55.3%	estimated ± 10.1 pp, low confidence
65	Grok 4.3	53.6%	estimated ± 11.6 pp, low confidence
66	Grok 4.20	53.3%	measured
67	GPT-5.2	52.9%	measured
68	GLM-5.3	52.7%	estimated ± 12.6 pp, low confidence
69	Grok 4.5	52.6%	measured
70	Claude Opus 4.6	52.6%	estimated ± 10.1 pp, low confidence
71	Ornith-1.5-35B-A3B	52.6%	estimated ± 10.1 pp, low confidence
72	Qwen3.8-27B	52.6%	estimated ± 10.1 pp, low confidence
73	Gemini 3 Flash	51.9%	estimated ± 12.6 pp, low confidence
74	Solar Pro 4	51.8%	estimated ± 10.1 pp, low confidence
75	Inkling	48.0%	estimated ± 10.1 pp, low confidence
76	MiMo-V2-Pro	47.1%	estimated ± 14.2 pp, low confidence
77	Kimi K2.5	47.0%	estimated ± 10.1 pp, low confidence
78	GPT-5.2-Codex	46.6%	estimated ± 14.2 pp, low confidence
79	Qwen 3.6 Max (preview)	46.2%	estimated ± 14.2 pp, low confidence
80	Hy3 Preview	45.8%	estimated ± 10.1 pp, low confidence
81	GLM-5.3-Flash	45.4%	estimated ± 12.6 pp, low confidence
82	MiniMax M2.7	45.2%	estimated ± 10.1 pp, low confidence
83	Nemotron 3 Ultra	45.2%	estimated ± 10.1 pp, low confidence
84	Gemini 3 Pro Deep Think	45.1%	measured
85	Ornith-1.5-9B	43.5%	estimated ± 10.1 pp, low confidence
86	Solar Open 2	43.2%	estimated ± 10.1 pp, low confidence
87	GLM-5.1	42.9%	estimated ± 10.1 pp, low confidence
88	Qwen3.5 397B	42.7%	estimated ± 11.6 pp, low confidence
89	Muse Spark	42.5%	measured
90	GLM-5	42.4%	estimated ± 10.1 pp, low confidence
91	Ternary Bonsai 2 27B	41.8%	estimated ± 10.1 pp, low confidence
92	A.X K2	41.4%	estimated ± 10.1 pp, low confidence
93	Quasar 438B	40.4%	estimated ± 14.2 pp, low confidence
94	Inkling-Small	40.1%	measured
95	GLM-5-Turbo	39.9%	estimated ± 14.2 pp, low confidence
96	Ling 3.0 Flash	39.8%	estimated ± 10.1 pp, low confidence
97	Apodex 1.1	39.2%	estimated ± 14.2 pp, low confidence
98	Apodex 1.1 Mini	39.2%	estimated ± 14.2 pp, low confidence
99	Qwen3.6-27B	38.8%	estimated ± 11.6 pp, low confidence
100	MAI-Thinking-1	38.0%	estimated ± 10.1 pp, low confidence
101	Claude Opus 4.5 Thinking	37.6%	measured
102	Ling 3.0 Flash FP8	37.6%	estimated ± 10.1 pp, low confidence
103	Kimi K2.5 (Reasoning)	37.5%	estimated ± 11.6 pp, low confidence
104	Kimi K2.7 Code	37.0%	estimated ± 14.2 pp, low confidence
105	Hy3	35.1%	estimated ± 14.2 pp, low confidence
106	Gemini 3.5 Flash-Lite	34.1%	estimated ± 12.6 pp, low confidence
107	K-EXAONE 2.0	33.9%	estimated ± 10.1 pp, low confidence
108	Claude Opus 4.5	33.6%	estimated ± 11.6 pp, low confidence
109	Ling 3.0 Flash VL	32.4%	estimated ± 14.2 pp, low confidence
110	Gemini 3 Pro	31.1%	measured
111	Qwen3.5-122B-A10B	31.0%	estimated ± 11.6 pp, low confidence
112	MiMo-V2-Omni	30.0%	estimated ± 14.2 pp, low confidence
113	Qwen3.5 Flash	29.8%	estimated ± 12.6 pp, low confidence
114	GPT-5.1-Codex	29.2%	estimated ± 14.2 pp, low confidence
115	GPT-5.1-Codex-Max	29.2%	estimated ± 14.2 pp, low confidence
116	MiMo-V2.5-Pro	28.9%	estimated ± 12.6 pp, low confidence
117	Mercury 2.5	28.7%	estimated ± 10.1 pp, low confidence
118	GLM-5V-Turbo	28.5%	estimated ± 14.2 pp, low confidence
119	Gemma 4 12B	28.4%	estimated ± 10.1 pp, low confidence
120	GPT-5.1	28.3%	estimated ± 5.7 pp, medium confidence
121	Qwen3.6-35B-A3B	27.2%	estimated ± 11.6 pp, low confidence
122	GPT-5 (high)	26.6%	estimated ± 14.2 pp, low confidence
123	GPT-5 (medium)	26.2%	estimated ± 14.2 pp, low confidence
124	Claude 4.1 Opus Thinking	26.2%	estimated ± 14.2 pp, low confidence
125	GLM-4.7	25.2%	estimated ± 11.6 pp, low confidence
126	Trinity-Large-Thinking	25.1%	estimated ± 10.1 pp, low confidence
127	Command A+	24.9%	estimated ± 14.2 pp, low confidence
128	Grok 4	24.9%	estimated ± 14.2 pp, low confidence
129	MiMo-V2.5	24.6%	estimated ± 12.6 pp, low confidence
130	Nemotron 3.5 Lightning 30B A3B NVFP4	24.3%	estimated ± 10.1 pp, low confidence
131	Qwen3.5-27B	23.9%	estimated ± 11.6 pp, low confidence
132	o3-pro	22.7%	estimated ± 14.2 pp, low confidence
133	Gemini 3.1 Flash-Lite	22.4%	estimated ± 12.6 pp, low confidence
134	Qwen3.5 397B (Reasoning)	21.3%	estimated ± 14.2 pp, low confidence
135	Nemotron 3 Nano Omni 30B A3B	20.9%	estimated ± 10.1 pp, low confidence
136	ZAYA1-8B	19.8%	estimated ± 10.1 pp, low confidence
137	MiniCPM5-2B	19.1%	estimated ± 10.1 pp, low confidence
138	GPT-5.4 mini	18.9%	measured
139	LongCat-Flash-Lite-Sparse	18.6%	estimated ± 10.1 pp, low confidence
140	Grok 4.1 Fast (Reasoning)	17.8%	estimated ± 14.2 pp, low confidence
141	o3	17.3%	estimated ± 14.2 pp, low confidence
142	Gemma 4 31B	16.2%	estimated ± 11.6 pp, low confidence
143	Qwen3.5-35B-A3B	15.5%	estimated ± 11.6 pp, low confidence
144	Step 3.7 Flash	15.2%	estimated ± 14.2 pp, low confidence
145	Trinity-Large-Preview	14.4%	estimated ± 10.1 pp, low confidence
146	Claude Sonnet 4.5 Thinking	13.6%	measured
147	Claude Sonnet 4.5	13.6%	measured
148	Claude 4.1 Opus	12.7%	estimated ± 14.2 pp, low confidence
149	MiMo-V2-Flash	12.3%	estimated ± 11.6 pp, low confidence
150	Mellum2-12B-A2.5B-Thinking	11.6%	estimated ± 10.1 pp, low confidence
151	ZAYA1-74B-Preview	11.5%	estimated ± 10.1 pp, low confidence
152	Grok 4 Fast (Reasoning)	11.1%	estimated ± 14.2 pp, low confidence
153	Muse Glimmer 30B	10.0%	estimated ± 14.2 pp, low confidence
154	Claude 4 Sonnet	8.2%	estimated ± 14.2 pp, low confidence
155	Gemini 2.5 Pro	7.8%	estimated ± 11.6 pp, low confidence
156	DeepSeek V3.2	7.1%	estimated ± 14.2 pp, low confidence
157	LLaDA2.2-mini	7.0%	estimated ± 10.1 pp, low confidence
158	Soofi S 30B-A3B	6.7%	estimated ± 10.1 pp, low confidence
159	Qwen3 Max	6.4%	estimated ± 14.2 pp, low confidence
160	Mellum2-12B-A2.5B-Instruct	6.1%	estimated ± 10.1 pp, low confidence
161	GPT-5.4 nano	5.7%	measured
162	K-Exaone	4.5%	estimated ± 14.2 pp, low confidence
163	Grok Code Fast 1	4.1%	estimated ± 14.2 pp, low confidence
164	DeepSeek V3.1	3.7%	estimated ± 14.2 pp, low confidence
165	DeepSeek V3.1 (Reasoning)	3.4%	estimated ± 14.2 pp, low confidence
166	MiniCPM5-1B	3.2%	estimated ± 10.1 pp, low confidence
167	LFM2.5-230M	3.1%	estimated ± 10.1 pp, low confidence
168	DeepSeek-R1	3.1%	estimated ± 14.2 pp, low confidence
169	Nemotron 3 Super 100B	2.8%	estimated ± 14.2 pp, low confidence
170	Kimi K2	2.7%	estimated ± 14.2 pp, low confidence
171	GPT-OSS 120B	1.8%	estimated ± 14.2 pp, low confidence
172	o1-preview	1.6%	estimated ± 14.2 pp, low confidence
173	Grok 4.1 Fast	1.6%	estimated ± 14.2 pp, low confidence
174	Mistral Small 4	1.6%	estimated ± 14.2 pp, low confidence
175	Mistral Small 4 (Reasoning)	1.6%	estimated ± 14.2 pp, low confidence
176	GLM-4.5-Air	1.5%	estimated ± 14.2 pp, low confidence
177	Ling 3.0 Tiny	1.5%	estimated ± 14.2 pp, low confidence
178	Llama 4 Maverick	0.9%	estimated ± 14.2 pp, low confidence
179	North Mini Code	0.9%	estimated ± 14.2 pp, low confidence
180	Gemini 2.5 Flash	0.9%	estimated ± 14.2 pp, low confidence
181	DeepSeek V3 0324	0.8%	estimated ± 14.2 pp, low confidence
182	Mistral Large 3	0.7%	estimated ± 14.2 pp, low confidence
183	Mistral Medium 3	0.6%	estimated ± 14.2 pp, low confidence
184	GPT-OSS 20B	0.6%	estimated ± 14.2 pp, low confidence
185	Nemotron 3 Nano 30B	0.6%	estimated ± 14.2 pp, low confidence
186	Sarvam 105B	0.5%	estimated ± 14.2 pp, low confidence
187	Claude 3 Opus	0.5%	estimated ± 14.2 pp, low confidence
188	GPT-4o	0.4%	estimated ± 14.2 pp, low confidence
189	LFM2.5-2.6B	0.4%	estimated ± 14.2 pp, low confidence
190	DeepSeek R1 Distill Qwen 32B	0.4%	estimated ± 14.2 pp, low confidence
191	Llama 4 Scout	0.4%	estimated ± 14.2 pp, low confidence
192	Gemini 1.5 Pro	0.3%	estimated ± 14.2 pp, low confidence
193	Solar Pro 3	0.3%	estimated ± 14.2 pp, low confidence
194	Qwen3-Omni-30B-A3B-Thinking	0.3%	estimated ± 14.2 pp, low confidence
195	Ultravox v0.6 Llama 3.3 70B	0.3%	estimated ± 14.2 pp, low confidence
196	Mistral Large 2	0.3%	estimated ± 14.2 pp, low confidence
197	Nemotron Ultra 253B	0.3%	estimated ± 14.2 pp, low confidence
198	Llama 3.1 405B	0.2%	estimated ± 14.2 pp, low confidence
199	LFM2.5-8B-A1B	0.2%	estimated ± 14.2 pp, low confidence
200	GPT-4 Turbo	0.2%	estimated ± 14.2 pp, low confidence
201	Solar Pro 2	0.2%	estimated ± 14.2 pp, low confidence
202	Nova Pro	0.2%	estimated ± 14.2 pp, low confidence
203	Qwen2.5 Coder 32B Instruct	0.2%	estimated ± 14.2 pp, low confidence
204	GPT-4o mini	0.2%	estimated ± 14.2 pp, low confidence
205	Sarvam 30B	0.1%	estimated ± 14.2 pp, low confidence
206	Celeris-1	0.1%	estimated ± 14.2 pp, low confidence
207	Exaone 4.0 32B	0.1%	estimated ± 14.2 pp, low confidence
208	Qwen3-Omni-30B-A3B-Instruct	0.1%	estimated ± 14.2 pp, low confidence
209	Phi-4	0.1%	estimated ± 14.2 pp, low confidence
210	Phi-4 Multimodal Instruct	0.1%	estimated ± 14.2 pp, low confidence
211	Claude 3 Haiku	0.1%	estimated ± 14.2 pp, low confidence
212	Gemini 1.0 Pro	0.1%	estimated ± 14.2 pp, low confidence
213	Exaone 4.0 1.2B	0.1%	estimated ± 14.2 pp, low confidence
214	Granite-4.0-H-1B	0.1%	estimated ± 14.2 pp, low confidence
215	Gemma 3 27B	0.0%	estimated ± 14.2 pp, low confidence
216	Granite-4.0-350M	0.0%	estimated ± 14.2 pp, low confidence
217	Granite-4.0-H-350M	0.0%	estimated ± 14.2 pp, low confidence
218	LFM2.5-VL-1.6B-Extract	0.0%	estimated ± 14.2 pp, low confidence
219	Gemma 4 26B A4B	0.0%	estimated ± 13.0 pp, low confidence
220	Claude 3.5 Sonnet	0.0%	estimated ± 11.6 pp, low confidence
221	Claude Haiku 4.5	0.0%	estimated ± 12.6 pp, low confidence
222	DeepSeek V3	0.0%	estimated ± 11.6 pp, low confidence
223	Gemma 4 E2B	0.0%	estimated ± 11.6 pp, low confidence
224	Gemma 4 E4B	0.0%	estimated ± 11.6 pp, low confidence
225	GLM-4.5	0.0%	estimated ± 12.6 pp, low confidence
226	GLM-4.6	0.0%	estimated ± 12.6 pp, low confidence
227	GPT-4.1	0.0%	estimated ± 11.6 pp, low confidence
228	GPT-4.1 mini	0.0%	estimated ± 11.6 pp, low confidence
229	GPT-4.1 nano	0.0%	estimated ± 11.6 pp, low confidence
230	Granite 4.2 30B	0.0%	estimated ± 11.6 pp, low confidence
231	Granite 4.2 3B	0.0%	estimated ± 11.6 pp, low confidence
232	Granite 4.2 8B	0.0%	estimated ± 11.6 pp, low confidence
233	Laguna M.1	0.0%	estimated ± 12.6 pp, low confidence
234	Laguna XS.2	0.0%	estimated ± 12.6 pp, low confidence
235	LFM2.5-VL-450M	0.0%	estimated ± 11.6 pp, low confidence
236	Ling 2.6 Flash	0.0%	estimated ± 11.6 pp, low confidence
237	Mistral Medium 3.5 128B	0.0%	estimated ± 12.6 pp, low confidence
238	o1	0.0%	estimated ± 11.6 pp, low confidence
239	o1-pro	0.0%	estimated ± 11.6 pp, low confidence
240	o3-mini	0.0%	estimated ± 11.6 pp, low confidence
241	Qwen3 235B 2507	0.0%	estimated ± 11.6 pp, low confidence

ARC-AGI-2 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General