llmrun
Reasoning

ARC-AGI-2 Leaderboard

ARC-AGI-2 is the second-generation ARC Prize puzzle set, rebuilt so that tasks solved by brute-force search or memorization were removed and every problem still requires inferring a novel rule from a handful of examples. It is markedly harder than the original ARC-AGI, and most models — open and closed alike — score near zero.

Source: epoch16 open models ranked+174 proprietaryData through Sep 2026

Open modelsAll models

All models ranked on ARC-AGI-2

Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.

#ModelScore
1GPT 6 Astra Max · proprietary
95.0%
2GPT 6 Astra (xhigh) · proprietary
93.3%
3GPT 5.6 Sol Max · proprietary
92.5%
4GPT 6 Astra (high) · proprietary
92.1%
5GPT 6 Astra (medium) · proprietary
92.1%
6Claude Opus 5 Max · proprietary
90.4%
7Claude Fable 5.1 (xhigh) · proprietary
90.0%
8Claude Fable 5.1 Max · proprietary
90.0%
9GPT 5.6 Sol (xhigh) · proprietary
90.0%
10Claude Fable 5 Max · proprietary
89.2%
11Claude Fable 5.1 (high) · proprietary
88.8%
12Claude Fable 5 (xhigh) · proprietary
88.3%
13Claude Opus 5 (high) · proprietary
88.3%
14Claude Fable 5 (high) · proprietary
87.5%
15Claude Fable 5.1 (medium) · proprietary
86.3%
16GPT 5.6 Sol (high) · proprietary
85.4%
17GPT 6 Astra (low) · proprietary
85.4%
18GPT 5.5 (xhigh) · proprietary
85.0%
19Gemini 3 Deep Think Preview · proprietary
84.6%
20Gemini 3.7 Flash (high) · proprietary
84.6%
21GPT 5.5 Pro (high) · proprietary
84.6%
22GPT 5.5 Pro (xhigh) · proprietary
84.2%
23GPT 5.6 Terra Max · proprietary
83.9%
24GPT 5.4 Pro (Mar 05, 2026, xhigh) · proprietary
83.3%
25GPT 5.5 (high) · proprietary
83.3%
26Claude Fable 5 (medium) · proprietary
82.5%
27Claude Fable 5.1 (low) · proprietary
78.3%
28Gemini 3.1 Pro Preview · proprietary
77.1%
29Claude Fable 5 (low) · proprietary
76.8%
30Claude Opus 4.7 Max · proprietary
75.8%
31GPT 5.6 Terra (xhigh) · proprietary
74.2%
32GPT 5.4 (Mar 05, 2026, xhigh) · proprietary
74.0%
33Claude Opus 4.8 (high) · proprietary
72.1%
34Gemini 3.5 Flash (high) · proprietary
72.1%
35Claude Opus 4.8 (medium) · proprietary
71.7%
36GPT 5.5 (medium) · proprietary
70.4%
37Claude Opus 4.6 (120K) · proprietary
69.2%
38Claude Opus 4.7 (high) · proprietary
68.3%
39Claude Opus 4.7 (medium) · proprietary
67.5%
40GPT 5.4 (Mar 05, 2026, high) · proprietary
67.5%
41GPT 5.6 Sol (medium) · proprietary
67.1%
42GPT 5.6 Terra (high) · proprietary
67.1%
43Grok 4.6 (xhigh) · proprietary
67.1%
44Grok 4.20 · proprietary
65.1%
45Grok 4.6 (high) · proprietary
65.1%
46Gemini 3.7 Flash (medium) · proprietary
63.7%
47Claude Opus 4.8 (low) · proprietary
62.2%
48Claude Opus 4.7 (low) · proprietary
62.1%
49DeepSeek V4 Flash 0731 · 304.2B
61.4%
50DeepSeek V4 Pro 0813 · 1650.5B
61.3%
51Grok 4.6 (medium) · proprietary
61.3%
52Claude Sonnet 4.6 (high) · proprietary
60.4%
53Gemini 3.6 Flash (high) · proprietary
60.4%
54Kimi K3 · 2779.9B
60.4%
55GPT 6 Astra None · proprietary
59.6%
56GPT 5.6 Luna Max · proprietary
59.5%
57Claude Sonnet 4.6 Max · proprietary
58.3%
58GPT 5.4 (Mar 05, 2026, medium) · proprietary
55.4%
59GPT 5.2 Pro (Dec 11, 2025, high) · proprietary
54.2%
60Gemini 3.7 Flash (low) · proprietary
52.9%
61GPT 5.2 (Dec 11, 2025, xhigh) · proprietary
52.9%
62Grok 4.5 (high) · proprietary
52.6%
63Grok 4.5 (medium) · proprietary
52.6%
64Gemini 3.6 Flash (medium) · proprietary
50.4%
65GPT 5.6 Luna (xhigh) · proprietary
47.6%
66GPT 5.2 (Dec 11, 2025, high) · proprietary
43.3%
67GPT 5.6 Sol (low) · proprietary
42.5%
68Inkling Small · 266.0B
40.1%
69GPT 5.2 Pro (Dec 11, 2025, medium) · proprietary
38.5%
70Claude Opus 4.5 (Nov 01, 2025, 64K) · proprietary
37.6%
71GPT 5.6 Terra (medium) · proprietary
37.5%
72Inkling · 952.4B
36.5%
73Gemini 3 Flash Preview · proprietary
33.6%
74GPT 5.5 (low) · proprietary
33.3%
75Grok 4.5 (low) · proprietary
33.1%
76Gemini 3 Pro Preview · proprietary
31.1%
77Gemini 3.6 Flash (low) · proprietary
30.4%
78GPT 5.6 Luna (high) · proprietary
29.3%
79GPT 5.4 (Mar 05, 2026, low) · proprietary
29.2%
80Grok 4.6 (low) · proprietary
27.6%
81GPT 5.2 (Dec 11, 2025, medium) · proprietary
26.7%
82Claude Opus 4.5 (Nov 01, 2025, 16K) · proprietary
22.8%
83GLM 5.2 · 753.3B
22.8%
84GPT 5.4 Mini (Mar 17, 2026, xhigh) · proprietary
18.9%
85GPT 5.6 Terra (low) · proprietary
18.8%
86GPT 5 Pro (Oct 06, 2025, unspecified) · proprietary
18.3%
87GPT 5.1 (Nov 13, 2025, high) · proprietary
17.6%
88Grok 4 (Jul 09) · proprietary
16.0%
89Claude Opus 4.5 (Nov 01, 2025, 8K) · proprietary
13.9%
90Claude Sonnet 4.5 (Sep 29, 2025, 32K) · proprietary
13.6%
91GPT 5.4 Mini (Mar 17, 2026, high) · proprietary
13.2%
92Kimi K2.5 · 1026.9B
11.8%
93Gemini 3.5 Flash Lite (high) · proprietary
10.3%
94GPT 5 (Aug 07, 2025, high) · proprietary
9.9%
95GPT 5.2 (Dec 11, 2025, low) · proprietary
9.7%
96Gemini 3.5 Flash (minimal) · proprietary
8.9%
97Claude Opus 4 (May 14, 2025, 16K) · proprietary
8.6%
98Claude Opus 4.5 (Nov 01, 2025) · proprietary
7.8%
99GPT 5 (Aug 07, 2025, medium) · proprietary
7.5%
100GPT 5.6 Luna (medium) · proprietary
7.4%
101Claude Sonnet 4.5 (Sep 29, 2025, 16K) · proprietary
6.9%
102Claude Sonnet 4.5 (Sep 29, 2025, 8K) · proprietary
6.9%
103GPT 5.1 (Nov 13, 2025, medium) · proprietary
6.5%
104O3 (Apr 16, 2025, high) · proprietary
6.5%
105Tiny Recursion Model · proprietary
6.3%
106O4 Mini (Apr 16, 2025, high) · proprietary
6.1%
107Claude Sonnet 4 (May 14, 2025, 16K) · proprietary
5.9%
108Claude Sonnet 4.5 (Sep 29, 2025, 1K) · proprietary
5.8%
109GPT 5.4 Nano (Mar 17, 2026, xhigh) · proprietary
5.7%
110Gemini 3.5 Flash Lite (medium) · proprietary
5.3%
111Grok 4 Fast · proprietary
5.3%
112GPT 5.6 Luna (low) · proprietary
5.1%
113Gemini 2.5 Pro (32K) · proprietary
4.9%
114GLM 5 · 753.9B
4.9%
115MiniMax M2.5 · 228.7B
4.9%
116O3 Pro (Jun 10, 2025, high) · proprietary
4.9%
117Claude Opus 4 (May 14, 2025, 8K) · proprietary
4.5%
118GPT 5 Mini (Aug 07, 2025, high) · proprietary
4.4%
119GPT 5.4 Mini (Mar 17, 2026, medium) · proprietary
4.4%
120Claude Haiku 4.5 (Oct 01, 2025, 32K) · proprietary
4.0%
121DeepSeek V3.2 · 685.4B
4.0%
122Gemini 2.5 Pro (16K) · proprietary
4.0%
123GPT 5 Mini (Aug 07, 2025, medium) · proprietary
4.0%
124Claude Sonnet 4.5 (Sep 29, 2025) · proprietary
3.8%
125GPT 5.4 Nano (Mar 17, 2026, high) · proprietary
3.6%
126O3 Mini (Jan 31, 2025, high) · proprietary
3.0%
127O3 (Apr 16, 2025, medium) · proprietary
3.0%
128Gemini 2.5 Pro (8K) · proprietary
2.9%
129Claude Haiku 4.5 (Oct 01, 2025, 16K) · proprietary
2.8%
130Gemini 3.6 Flash (minimal) · proprietary
2.6%
131GPT 5 Nano (Aug 07, 2025, high) · proprietary
2.6%
132Gemini 2.5 Flash Preview (May 20, 23K) · proprietary
2.5%
133O4 Mini (Apr 16, 2025, medium) · proprietary
2.4%
134Gemini 2.5 Flash Preview (May 20, 1K) · proprietary
2.2%
135Claude Sonnet 4 (May 14, 2025, 8K) · proprietary
2.1%
136Gemini 2.5 Flash Preview (May 20, 8K) · proprietary
2.1%
137O3 Mini (Jan 31, 2025, medium) · proprietary
2.1%
138O3 Pro (Jun 10, 2025, low) · proprietary
2.1%
139O3 (Apr 16, 2025, low) · proprietary
2.0%
140Gemini 2.5 Flash Preview (May 20, 16K) · proprietary
2.0%
141GPT 5 (Aug 07, 2025, low) · proprietary
1.9%
142GPT 5.1 (Nov 13, 2025, low) · proprietary
1.9%
143GPT 5.4 Nano (Mar 17, 2026, medium) · proprietary
1.9%
144O3 Pro (Jun 10, 2025, medium) · proprietary
1.9%
145Gemini 2.5 Flash Preview (May 20) · proprietary
1.7%
146Claude Haiku 4.5 (Oct 01, 2025, 8K) · proprietary
1.7%
147GPT 5 Mini (Aug 07, 2025, minimal) · proprietary
1.7%
148O4 Mini (Apr 16, 2025, low) · proprietary
1.7%
149Gemini 3.5 Flash Lite (low) · proprietary
1.5%
150GPT 5.4 Nano (Mar 17, 2026, low) · proprietary
1.5%
151DeepSeek R1 · 684.5B
1.3%
152Gemini 2.0 Flash 001 · proprietary
1.3%
153Claude Opus 4 (May 14, 2025) · proprietary
1.3%
154Claude Sonnet 4 (May 14, 2025) · proprietary
1.3%
155Codex Mini (May 16, 2025) · proprietary
1.3%
156Claude Haiku 4.5 (Oct 01, 2025, 1K) · proprietary
1.3%
157Claude Haiku 4.5 (Oct 01, 2025) · proprietary
1.3%
158Qwen3 235B A22B Instruct 2507 · 235.1B
1.3%
159DeepSeek R1 0528 · 684.5B
1.1%
160GPT 5.4 Mini (Mar 17, 2026, low) · proprietary
1.1%
161Claude 3.7 Sonnet (Feb 19, 2025, 8K) · proprietary
0.9%
162GPT 5 Nano (Aug 07, 2025, medium) · proprietary
0.9%
163Claude Sonnet 4 (May 14, 2025, 1K) · proprietary
0.9%
164Gemini 3.5 Flash Lite (minimal) · proprietary
0.8%
165GPT 5 Mini (Aug 07, 2025, low) · proprietary
0.8%
166GPT 5.2 (Dec 11, 2025, unspecified) · proprietary
0.8%
167O1 Mini (Sep 12, 2024, unspecified) · proprietary
0.8%
168Gemini 1.5 Pro 002 · proprietary
0.8%
169GPT 4.5 Preview (Feb 27, 2025) · proprietary
0.8%
170Claude 3.7 Sonnet (Feb 19, 2025, 16K) · proprietary
0.7%
171GPT 4.1 (Apr 14, 2025) · proprietary
0.4%
172GPT 5.1 2025 11.13 None · proprietary
0.4%
173Grok 3 Mini (low) · proprietary
0.4%
174Claude 3.7 Sonnet (Feb 19, 2025, 1K) · proprietary
0.4%
175Claude 3.7 Sonnet (Feb 19, 2025) · proprietary
0.0%
176Claude Opus 4 (May 14, 2025, 1K) · proprietary
0.0%
177Gemini 2.5 Pro Preview (Jun 05, 1K) · proprietary
0.0%
178GPT 4.1 Mini (Apr 14, 2025) · proprietary
0.0%
179GPT 4.1 Nano (Apr 14, 2025) · proprietary
0.0%
180GPT 4o (Nov 20, 2024) · proprietary
0.0%
181GPT 4o Mini (Jul 18, 2024) · proprietary
0.0%
182GPT 5 (Aug 07, 2025, minimal) · proprietary
0.0%
183GPT 5 Nano (Aug 07, 2025, low) · proprietary
0.0%
184GPT 5 Nano (Aug 07, 2025, minimal) · proprietary
0.0%
185Grok 3 · proprietary
0.0%
186Llama 4 Maverick 17B 128E Instruct · 401.6B
0.0%
187Llama 4 Scout 17B 16E Instruct · 108.6B
0.0%
188Magistral Medium 2506 · proprietary
0.0%
189Magistral Small 2506 · 23.6B
0.0%
190O3 Mini (Jan 31, 2025, low) · proprietary
0.0%

Score vs model size

Which models give the most quality for their size — the ones worth running locally.

100B1Tmodel size (log scale) →61.4%0.0%DeepSeek V4 Pro 0813 · 1.7T · 61.3%Kimi K3 · 2.8T · 60.4%Inkling · 952B · 36.5%GLM 5.2 · 753B · 22.8%Kimi K2.5 · 1T · 11.8%GLM 5 · 754B · 4.9%DeepSeek V3.2 · 685B · 4.0%DeepSeek R1 · 684B · 1.3%Qwen3 235B A22B Instruct 2507 · 235B · 1.3%DeepSeek R1 0528 · 685B · 1.1%Llama 4 Maverick 17B 128E Instruct · 402B · 0.0%Llama 4 Scout 17B 16E Instruct · 109B · 0.0%Magistral Small 2506 · 24B · 0.0%Magistral Small 2506MiniMax M2.5 · 229B · 4.9%MiniMax M2.5Inkling Small · 266B · 40.1%Inkling SmallDeepSeek V4 Flash 0731 · 304B · 61.4%DeepSeek V4 Flash 0731
Each dot is a model. Up = higher score, left = smaller (easier to run locally). The dashed line marks the efficiency frontier — the best score you can get at each size or smaller.
  • Magistral Small 2506, 24B, score 0.0% — on the efficiency frontier (best score at its size or smaller).
  • MiniMax M2.5, 229B, score 4.9% — on the efficiency frontier (best score at its size or smaller).
  • Inkling Small, 266B, score 40.1% — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Flash 0731, 304B, score 61.4% — on the efficiency frontier (best score at its size or smaller).

ARC-AGI-2: frequently asked questions

What is the best open LLM on ARC-AGI-2?
DeepSeek V4 Flash 0731 is the top open model on ARC-AGI-2, scoring 61.4%. Among all models tested — including proprietary ones — it ranks #49. The top model overall is GPT 6 Astra Max (OpenAI) at 95.0%.
What's the best ARC-AGI-2 model you can run on a 24 GB GPU?
Magistral Small 2506 is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 13 GB), scoring 0.0% on ARC-AGI-2.
Can open models match proprietary models on ARC-AGI-2?
Not quite on ARC-AGI-2: the strongest proprietary model (GPT 6 Astra Max) scores 95.0%, ahead of the best open model (DeepSeek V4 Flash 0731) at 61.4% — but you can run the open one yourself.

Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.