Knowledge & reasoning

ARC-AGI-3 Provider Adapter leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on ARC-AGI-3 Provider Adapter is 98.6% by GPT-6 Astra. 107 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	98.6%	estimated ± 7.5 pp, low confidence
2	Gemini 4 Carbon	98.6%	estimated ± 7.5 pp, low confidence
3	Claude Fable 5.1	98.6%	estimated ± 7.5 pp, low confidence
4	GPT-6 Astra	98.6%	measured
5	Claude Fable 5	98.3%	estimated ± 7.5 pp, medium confidence
6	Claude Sonnet 5.5	98.1%	estimated ± 7.5 pp, medium confidence
7	Gemini 4 Argon	97.3%	measured
8	Muse Wonder Full	96.9%	estimated ± 7.5 pp, medium confidence
9	GPT-6.1 Sol	96.2%	measured
10	Kimi K3.1 Pro	94.3%	estimated ± 7.5 pp, medium confidence
11	Claude Opus 5	75.7%	estimated ± 4.3 pp, high confidence
12	Muse Wonder	72.2%	estimated ± 7.5 pp, medium confidence
13	MiMo-V2.6-Pro	59.3%	estimated ± 7.5 pp, medium confidence
14	Muse Spark 1.3	44.9%	estimated ± 7.5 pp, medium confidence
15	Gemini 3.8 Flash	35.0%	measured
16	MiniMax M3.1 Flash	31.5%	estimated ± 7.5 pp, medium confidence
17	GPT-5.6 Sol	31.0%	estimated ± 4.3 pp, high confidence
18	Gemini 3.7 Flash	29.7%	estimated ± 7.5 pp, medium confidence
19	GPT-6 Sol	23.0%	measured
20	Kimi K3	16.3%	estimated ± 7.5 pp, medium confidence
21	Step 5 Preview	12.8%	estimated ± 7.5 pp, medium confidence
22	GPT-6.1 Terra	10.7%	estimated ± 7.5 pp, medium confidence
23	Muse Spark 1.1	10.7%	estimated ± 7.5 pp, medium confidence
24	Grok 4.6	9.9%	estimated ± 4.3 pp, high confidence
25	Muse Spark 1.2	7.3%	estimated ± 7.5 pp, medium confidence
26	Claude Opus 4.8	7.2%	estimated ± 4.3 pp, high confidence
27	Grok 4.7	7.0%	measured
28	Ling 3.0 Flash Sante	7.0%	estimated ± 13.7 pp, low confidence
29	Muse Spark Open	3.9%	estimated ± 7.5 pp, medium confidence
30	GPT-5.6 Terra	3.9%	estimated ± 4.3 pp, high confidence
31	Qwen3.8 Max Preview	3.4%	estimated ± 7.5 pp, medium confidence
32	Gemini 3.5 Flash	3.2%	estimated ± 7.5 pp, medium confidence
33	GPT-5.3 Codex	3.1%	estimated ± 7.5 pp, medium confidence
34	GLM-5.3	3.1%	estimated ± 7.5 pp, medium confidence
35	Claude Sonnet 5	2.9%	estimated ± 7.5 pp, medium confidence
36	GLM-5.2	2.9%	estimated ± 7.5 pp, medium confidence
37	DeepSeek V4 Pro 0813	2.9%	estimated ± 7.5 pp, medium confidence
38	Gemini 3.6 Flash	2.9%	estimated ± 7.5 pp, medium confidence
39	Muse Spark	2.9%	estimated ± 7.5 pp, medium confidence
40	Qwen3.7 Max	2.8%	estimated ± 7.5 pp, medium confidence
41	Claude Opus 4.6 (Adaptive)	2.8%	estimated ± 7.5 pp, medium confidence
42	Claude Haiku 5.5	2.8%	measured
43	Gemini 3 Pro	2.8%	estimated ± 7.5 pp, medium confidence
44	Ling 3.1 Flash	2.8%	estimated ± 7.5 pp, medium confidence
45	DeepSeek V4.1 Flash 1017	2.8%	estimated ± 7.5 pp, medium confidence
46	MiniMax M3	2.8%	estimated ± 7.5 pp, medium confidence
47	DeepSeek V4 Flash 0731	2.8%	estimated ± 7.5 pp, medium confidence
48	Qwen3.8-Flash-Next	2.8%	estimated ± 7.5 pp, low confidence
49	GPT-5.2	2.8%	estimated ± 7.5 pp, low confidence
50	Kimi K2.6	2.8%	estimated ± 7.5 pp, low confidence
51	Grok 4.3	2.8%	estimated ± 7.5 pp, low confidence
52	GPT-5.2-Codex	2.8%	estimated ± 7.5 pp, low confidence
53	MiMo-V2.5-Pro	2.8%	estimated ± 7.5 pp, low confidence
54	Qwen3.7 Plus	2.8%	estimated ± 7.5 pp, low confidence
55	MiMo-V2.6-Flash	2.8%	estimated ± 7.5 pp, low confidence
56	Kimi K2.7 Code	2.8%	estimated ± 7.5 pp, low confidence
57	Mistral Large 4	2.8%	estimated ± 7.5 pp, low confidence
58	Beam 1.0 501B-A23B	2.8%	estimated ± 7.5 pp, low confidence
59	Apodex 1.1	2.8%	estimated ± 7.5 pp, low confidence
60	Apodex 1.1 Mini	2.8%	estimated ± 7.5 pp, low confidence
61	Qwen3.8-27B	2.8%	estimated ± 7.5 pp, low confidence
62	Hy3	2.8%	estimated ± 7.5 pp, low confidence
63	Hy3 Preview	2.8%	estimated ± 7.5 pp, low confidence
64	Inkling-Small	2.8%	estimated ± 7.5 pp, low confidence
65	Inkling	2.8%	estimated ± 7.5 pp, low confidence
66	Qwen 3.6 Max (preview)	2.8%	estimated ± 7.5 pp, low confidence
67	Kimi K2.5	2.8%	estimated ± 7.5 pp, low confidence
68	Kimi K2.5 (Reasoning)	2.8%	estimated ± 7.5 pp, low confidence
69	Nemotron 3.5 Super VL 120B A12B	2.8%	estimated ± 7.5 pp, low confidence
70	Nemotron 3.5 Super VL 120B A12B NVFP4	2.8%	estimated ± 7.5 pp, low confidence
71	MiMo-V2-Pro	2.8%	estimated ± 7.5 pp, low confidence
72	Claude Opus 4.5 Thinking	2.8%	estimated ± 7.5 pp, low confidence
73	GLM-5.1	2.8%	estimated ± 7.5 pp, low confidence
74	Solar Pro 4	2.8%	estimated ± 7.5 pp, low confidence
75	A.X K2	2.8%	estimated ± 7.5 pp, low confidence
76	MiniMax M2.7	2.8%	estimated ± 7.5 pp, low confidence
77	GLM-5	2.8%	estimated ± 7.5 pp, low confidence
78	Qwen3.5 397B	2.8%	estimated ± 7.5 pp, low confidence
79	Qwen3.5 397B (Reasoning)	2.8%	estimated ± 7.5 pp, low confidence
80	GPT-5.1	2.8%	estimated ± 7.5 pp, low confidence
81	GPT-5 (high)	2.8%	estimated ± 7.5 pp, low confidence
82	Nemotron 3 Ultra	2.8%	estimated ± 7.5 pp, low confidence
83	GPT-5.4 nano	2.8%	estimated ± 7.5 pp, low confidence
84	GPT-5.4 mini	2.8%	estimated ± 7.5 pp, low confidence
85	GLM-5-Turbo	2.8%	estimated ± 7.5 pp, low confidence
86	Qwen3.6 Plus	2.8%	estimated ± 7.5 pp, low confidence
87	GLM-4.7	2.8%	estimated ± 7.5 pp, low confidence
88	Grok 4	2.8%	estimated ± 7.5 pp, low confidence
89	GPT-5.1-Codex	2.8%	estimated ± 7.5 pp, low confidence
90	GPT-5.1-Codex-Max	2.8%	estimated ± 7.5 pp, low confidence
91	GPT-5 (medium)	2.8%	estimated ± 7.5 pp, low confidence
92	Qwen3.5-122B-A10B	2.8%	estimated ± 7.5 pp, low confidence
93	Nemotron 3.5 Bolt 74B-A9B	2.8%	estimated ± 7.5 pp, low confidence
94	Qwen3.5-27B	2.8%	estimated ± 7.5 pp, low confidence
95	Ling 3.0 Flash	2.8%	estimated ± 7.5 pp, low confidence
96	Gemma 4 31B	2.8%	estimated ± 7.5 pp, low confidence
97	Qwen3.6-27B	2.8%	estimated ± 7.5 pp, low confidence
98	Gemini 2.5 Pro	2.8%	estimated ± 7.5 pp, low confidence
99	Qwen3.6-35B-A3B	2.8%	estimated ± 7.5 pp, low confidence
100	MiMo-V2-Omni	2.8%	estimated ± 7.5 pp, low confidence
101	Ling 3.0 Flash VL	2.8%	estimated ± 7.5 pp, low confidence
102	Muse Glimmer 30B	2.8%	estimated ± 7.5 pp, low confidence
103	Step 3.7 Flash	2.8%	estimated ± 7.5 pp, low confidence
104	Qwen3.5-35B-A3B	2.8%	estimated ± 7.5 pp, low confidence
105	o3	2.8%	estimated ± 7.5 pp, low confidence
106	Claude Opus 4.6	2.5%	estimated ± 4.3 pp, high confidence
107	GPT-5.5	2.1%	estimated ± 4.3 pp, high confidence
108	Gemini 3.1 Pro	2.1%	estimated ± 4.3 pp, high confidence
109	Grok 4.5	1.5%	estimated ± 4.3 pp, high confidence
110	GPT-5.4	1.0%	estimated ± 4.3 pp, high confidence
111	Claude Opus 4.7	0.9%	estimated ± 4.3 pp, high confidence
112	Claude Opus 4.7 (Adaptive)	0.9%	estimated ± 4.3 pp, high confidence
113	GPT-5.6 Luna	0.9%	estimated ± 4.3 pp, high confidence
114	GPT-6 Luna	0.6%	measured
115	Grok 4.20	0.4%	estimated ± 4.3 pp, medium confidence

ARC-AGI-3 Provider Adapter leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General