benchgap
Knowledge & reasoning

ARC-AGI-3 leaderboard

As of 2026-11-21, the highest measured score on ARC-AGI-3 is 70.3% by GPT-6.2 Astra. 234 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra70.3%measured
2Claude Fable 5.265.7%measured
3Gemini 4 Carbon63.9%measured
4GPT-6 Astra62.7%measured
5GPT-6.2 Sol54.9%measured
6GPT-6.1 Sol52.7%measured
7Muse Wonder Full51.4%estimated ± 9.8 pp, medium confidence
8Gemini 4 Argon46.3%measured
9Claude Opus 5.538.8%estimated ± 7.2 pp, medium confidence
10Claude Fable 5.130.6%estimated ± 7.2 pp, medium confidence
11Claude Opus 530.2%measured
12Claude Mythos 526.5%estimated ± 7.2 pp, medium confidence
13Claude Sonnet 5.522.3%estimated ± 7.2 pp, medium confidence
14Muse Wonder16.7%estimated ± 9.8 pp, medium confidence
15Hy416.4%estimated ± 8.3 pp, medium confidence
16Muse Spark 1.114.3%estimated ± 7.2 pp, medium confidence
17Qwen4 Max13.6%estimated ± 7.2 pp, medium confidence
18Kimi K3.1 Pro12.8%estimated ± 7.2 pp, medium confidence
19Claude Fable 511.4%estimated ± 8.6 pp, low confidence
20Sakana Fugu-Ultra11.4%estimated ± 7.2 pp, medium confidence
21Gemini 3.8 Flash10.4%measured
22Pareto 26.910.2%estimated ± 7.2 pp, medium confidence
23Sakana Fugu8.3%estimated ± 7.2 pp, medium confidence
24GPT-5.6 Sol7.8%measured
25Gemini 3.7 Flash7.6%estimated ± 8.6 pp, low confidence
26Gemini 3 Pro7.6%estimated ± 8.6 pp, low confidence
27ERNIE 5.5 Flash Preview7.5%estimated ± 7.2 pp, medium confidence
28Apodex 1.17.5%estimated ± 8.3 pp, medium confidence
29MiMo-V2.6-Pro7.5%estimated ± 9.8 pp, medium confidence
30Ornith-1.5-397B6.0%estimated ± 7.2 pp, medium confidence
31Ling 3.0 Flash Sante5.6%estimated ± 11.8 pp, low confidence
32Gemini 3.6 Flash5.5%estimated ± 8.6 pp, low confidence
33Qwen3.8 Max5.3%estimated ± 7.2 pp, medium confidence
34Kimi K35.2%estimated ± 7.2 pp, medium confidence
35Hy4 preview5.1%estimated ± 7.2 pp, medium confidence
36MiniMax M3.1 Flash5.1%estimated ± 7.2 pp, medium confidence
37Claude Sonnet 55.0%estimated ± 7.2 pp, medium confidence
38Claude Opus 4.6 (Adaptive)5.0%estimated ± 8.6 pp, low confidence
39GPT-5.5 Pro4.9%estimated ± 7.2 pp, medium confidence
40Hy4 flash4.8%estimated ± 8.3 pp, medium confidence
41Muse Spark4.8%estimated ± 7.2 pp, medium confidence
42GPT-5.4 Pro4.7%estimated ± 7.2 pp, medium confidence
43GPT-6 Sol4.6%measured
44Gemini 3 Flash3.6%estimated ± 8.6 pp, low confidence
45GLM-5.23.5%estimated ± 7.2 pp, medium confidence
46Muse Spark 1.33.4%estimated ± 9.8 pp, medium confidence
47Muse Spark 1.22.8%estimated ± 8.6 pp, low confidence
48Grok 4.62.1%measured
49Grok 4.71.8%measured
50Beam 1.0 501B-A23B1.7%estimated ± 7.2 pp, medium confidence
51Beam 1.0 501B-A23B FP81.7%estimated ± 7.2 pp, medium confidence
52Qwen3.8-Flash-Next1.7%estimated ± 7.2 pp, medium confidence
53Claude Opus 4.81.5%measured
54Beam 1.0 501B-A23B NVFP41.4%estimated ± 7.2 pp, medium confidence
55GPT-5.21.3%estimated ± 3.1 pp, low confidence
56GPT-6.1 Terra1.3%estimated ± 9.8 pp, medium confidence
57Muse Spark Open1.3%estimated ± 9.8 pp, medium confidence
58Qwen3.8 Max Preview1.3%estimated ± 9.8 pp, medium confidence
59GPT-5.3 Codex1.3%estimated ± 9.8 pp, medium confidence
60GPT-5.2-Codex1.3%estimated ± 9.8 pp, medium confidence
61MiMo-V2.6-Flash1.3%estimated ± 9.8 pp, medium confidence
62Kimi K2.7 Code1.3%estimated ± 9.8 pp, medium confidence
63Apodex 1.1 Mini1.3%estimated ± 9.8 pp, medium confidence
64Hy31.3%estimated ± 9.8 pp, medium confidence
65Qwen 3.6 Max (preview)1.3%estimated ± 9.8 pp, medium confidence
66MiMo-V2-Pro1.3%estimated ± 9.8 pp, medium confidence
67GLM-5-Turbo1.3%estimated ± 9.8 pp, medium confidence
68GPT-5.1-Codex1.3%estimated ± 9.8 pp, medium confidence
69GPT-5.1-Codex-Max1.3%estimated ± 9.8 pp, medium confidence
70Celeris-11.3%estimated ± 9.8 pp, low confidence
71Claude 3.5 Haiku1.3%estimated ± 9.8 pp, low confidence
72Claude 3 Haiku1.3%estimated ± 9.8 pp, low confidence
73Claude 3 Opus1.3%estimated ± 9.8 pp, low confidence
74Claude 4 Opus1.3%estimated ± 9.8 pp, low confidence
75Claude 4 Opus Thinking1.3%estimated ± 9.8 pp, low confidence
76Claude 4 Sonnet1.3%estimated ± 9.8 pp, low confidence
77Command A+1.3%estimated ± 9.8 pp, low confidence
78DeepSeek-R11.3%estimated ± 9.8 pp, low confidence
79DeepSeek R1 Distill Qwen 32B1.3%estimated ± 9.8 pp, low confidence
80DeepSeek V31.3%estimated ± 9.8 pp, low confidence
81DeepSeek V3 03241.3%estimated ± 9.8 pp, low confidence
82DeepSeek V3.11.3%estimated ± 9.8 pp, low confidence
83DeepSeek V3.21.3%estimated ± 9.8 pp, low confidence
84Devstral Small1.3%estimated ± 9.8 pp, low confidence
85Exaone 4.0 1.2B1.3%estimated ± 9.8 pp, low confidence
86Exaone 4.0 32B1.3%estimated ± 9.8 pp, low confidence
87Gemini 1.0 Pro1.3%estimated ± 9.8 pp, low confidence
88Gemini 1.5 Flash1.3%estimated ± 9.8 pp, low confidence
89Gemini 1.5 Pro1.3%estimated ± 9.8 pp, low confidence
90Gemini 2.0 Flash1.3%estimated ± 9.8 pp, low confidence
91Gemini 2.5 Flash1.3%estimated ± 9.8 pp, low confidence
92Gemini 2.5 Flash-Lite1.3%estimated ± 9.8 pp, low confidence
93Gemini 2.5 Pro1.3%estimated ± 9.8 pp, medium confidence
94Gemma 3 27B1.3%estimated ± 9.8 pp, low confidence
95Gemma 4 E2B1.3%estimated ± 9.8 pp, low confidence
96Gemma 4 E4B1.3%estimated ± 9.8 pp, low confidence
97GLM-4.5-Air1.3%estimated ± 9.8 pp, low confidence
98GLM-5V-Turbo1.3%estimated ± 9.8 pp, low confidence
99GPT-4.1 mini1.3%estimated ± 9.8 pp, low confidence
100GPT-4.1 nano1.3%estimated ± 9.8 pp, low confidence
101GPT-4 Turbo1.3%estimated ± 9.8 pp, low confidence
102GPT-4o1.3%estimated ± 9.8 pp, low confidence
103GPT-4o mini1.3%estimated ± 9.8 pp, low confidence
104GPT-OSS 120B1.3%estimated ± 9.8 pp, medium confidence
105GPT-OSS 20B1.3%estimated ± 9.8 pp, low confidence
106Granite-4.0-350M1.3%estimated ± 9.8 pp, low confidence
107Granite-4.0-H-1B1.3%estimated ± 9.8 pp, low confidence
108Granite-4.0-H-350M1.3%estimated ± 9.8 pp, low confidence
109Granite 4.2 30B1.3%estimated ± 9.8 pp, low confidence
110Granite 4.2 3B1.3%estimated ± 9.8 pp, low confidence
111Granite 4.2 8B1.3%estimated ± 9.8 pp, low confidence
112Grok 4.1 Fast1.3%estimated ± 9.8 pp, low confidence
113Grok 4 Fast (Reasoning)1.3%estimated ± 9.8 pp, medium confidence
114Grok Code Fast 11.3%estimated ± 9.8 pp, low confidence
115K-Exaone1.3%estimated ± 9.8 pp, low confidence
116Kimi K21.3%estimated ± 9.8 pp, low confidence
117LFM2.5-2.6B1.3%estimated ± 9.8 pp, low confidence
118LFM2.5-8B-A1B1.3%estimated ± 9.8 pp, low confidence
119LFM2.5-VL-1.6B-Extract1.3%estimated ± 9.8 pp, low confidence
120Ling 2.6 Flash1.3%estimated ± 9.8 pp, low confidence
121Ling 3.0 Flash VL1.3%estimated ± 9.8 pp, medium confidence
122Ling 3.0 Tiny1.3%estimated ± 9.8 pp, low confidence
123Llama 3.1 405B1.3%estimated ± 9.8 pp, low confidence
124Llama 3.1 70B1.3%estimated ± 9.8 pp, low confidence
125Llama 3.1 8B1.3%estimated ± 9.8 pp, low confidence
126Llama 3.3 70B1.3%estimated ± 9.8 pp, low confidence
127Llama 4 Maverick1.3%estimated ± 9.8 pp, low confidence
128Llama 4 Scout1.3%estimated ± 9.8 pp, low confidence
129Magistral Medium1.3%estimated ± 9.8 pp, low confidence
130Magistral Small1.3%estimated ± 9.8 pp, low confidence
131Mercury 2.51.3%estimated ± 9.8 pp, low confidence
132MiMo-V2-Flash1.3%estimated ± 9.8 pp, low confidence
133MiMo-V2-Omni1.3%estimated ± 9.8 pp, medium confidence
134MiniCPM5-2B1.3%estimated ± 9.8 pp, low confidence
135Mistral Large 21.3%estimated ± 9.8 pp, low confidence
136Mistral Large 31.3%estimated ± 9.8 pp, low confidence
137Mistral Medium 31.3%estimated ± 9.8 pp, low confidence
138Mistral Medium 3.5 128B1.3%estimated ± 9.8 pp, low confidence
139Mistral Small 3.11.3%estimated ± 9.8 pp, low confidence
140Mistral Small 41.3%estimated ± 9.8 pp, low confidence
141Muse Glimmer 30B1.3%estimated ± 9.8 pp, medium confidence
142Nemotron 3 Nano 30B1.3%estimated ± 9.8 pp, low confidence
143Nemotron 3 Nano Omni 30B A3B1.3%estimated ± 9.8 pp, low confidence
144Nemotron Ultra 253B1.3%estimated ± 9.8 pp, low confidence
145North Mini Code1.3%estimated ± 9.8 pp, low confidence
146Nova Pro1.3%estimated ± 9.8 pp, low confidence
147o1-mini1.3%estimated ± 9.8 pp, low confidence
148o3-mini1.3%estimated ± 9.8 pp, low confidence
149Phi-41.3%estimated ± 9.8 pp, low confidence
150Phi-4 Multimodal Instruct1.3%estimated ± 9.8 pp, low confidence
151Phi-4-reasoning1.3%estimated ± 9.8 pp, low confidence
152Quasar 438B1.3%estimated ± 9.8 pp, low confidence
153Qwen2.5 Coder 32B Instruct1.3%estimated ± 9.8 pp, low confidence
154Qwen3.5-27B1.3%estimated ± 9.8 pp, medium confidence
155Qwen3.5-35B-A3B1.3%estimated ± 9.8 pp, medium confidence
156Qwen3.5-4B1.3%estimated ± 9.8 pp, low confidence
157Qwen3-Omni-30B-A3B-Instruct1.3%estimated ± 9.8 pp, low confidence
158Qwen3-Omni-30B-A3B-Thinking1.3%estimated ± 9.8 pp, low confidence
159QwQ-32B1.3%estimated ± 9.8 pp, low confidence
160Sarvam 105B1.3%estimated ± 9.8 pp, low confidence
161Sarvam 30B1.3%estimated ± 9.8 pp, low confidence
162Solar Pro 21.3%estimated ± 9.8 pp, low confidence
163Solar Pro 31.3%estimated ± 9.8 pp, low confidence
164Step 3.7 Flash1.3%estimated ± 9.8 pp, medium confidence
165Trinity-Large-Preview1.3%estimated ± 9.8 pp, low confidence
166Trinity-Large-Thinking1.3%estimated ± 9.8 pp, low confidence
167Ultravox v0.6 Llama 3.3 70B1.3%estimated ± 9.8 pp, low confidence
168MiMo-V2.5-Pro1.2%estimated ± 7.2 pp, medium confidence
169Qwen3.6 Plus1.2%estimated ± 8.6 pp, low confidence
170Nemotron 3.5 Super VL 120B A12B0.8%estimated ± 7.2 pp, medium confidence
171GPT-5.6 Terra0.8%measured
172Inkling-Small0.8%estimated ± 7.2 pp, medium confidence
173Nemotron 3.5 Super VL 120B A12B NVFP40.8%estimated ± 7.2 pp, low confidence
174Qwen3.8-27B0.7%estimated ± 7.2 pp, low confidence
175Inkling 2 Mini0.7%estimated ± 7.2 pp, low confidence
176Inkling0.6%estimated ± 7.2 pp, low confidence
177Claude Opus 4.60.5%measured
178Solar Open 20.5%estimated ± 7.2 pp, low confidence
179Claude Haiku 5.50.5%measured
180Nemotron 3.5 Bolt 74B-A9B BF160.5%estimated ± 7.2 pp, low confidence
181GPT-5.50.4%measured
182Gemini 3.1 Pro0.4%measured
183Claude Sonnet 4.5 Thinking0.4%estimated ± 8.6 pp, low confidence
184GPT-5.4 mini0.4%estimated ± 7.2 pp, low confidence
185Grok 4.50.3%measured
186Inkling 2 Small0.3%measured
187Nemotron 3 Ultra0.3%estimated ± 7.2 pp, low confidence
188Ornith-1.5-35B-A3B0.2%estimated ± 7.2 pp, low confidence
189Nemotron 3.5 Bolt 74B-A9B0.2%estimated ± 7.2 pp, low confidence
190GPT-5.40.2%measured
191Claude Opus 4.70.2%measured
192GPT-5.6 Luna0.2%measured
193GPT-5.4 nano0.2%estimated ± 7.2 pp, low confidence
194Claude Opus 4.5 Thinking0.1%estimated ± 8.6 pp, low confidence
195GPT-6 Luna0.1%measured
196Grok 4.200.1%measured
197Ornith-1.5-9B0.1%estimated ± 7.2 pp, low confidence
198Gemma 4 31B0.0%estimated ± 7.2 pp, low confidence
199Spark-X2.6-4B0.0%estimated ± 7.2 pp, low confidence
200Nemotron 3.5 Lightning 30B A3B NVFP40.0%estimated ± 7.2 pp, low confidence
201Gemma 4 26B A4B0.0%estimated ± 7.2 pp, low confidence
202Qwen3.5 397B0.0%estimated ± 9.3 pp, low confidence
203Qwen3.6-27B0.0%estimated ± 9.3 pp, low confidence
204Gemma 4 12B0.0%estimated ± 7.2 pp, low confidence
205Qwen3.5-122B-A10B0.0%estimated ± 9.3 pp, low confidence
206Qwen3.6-35B-A3B0.0%estimated ± 9.3 pp, low confidence
207A.X K20.0%estimated ± 8.6 pp, low confidence
208Agents-A10.0%estimated ± 8.3 pp, low confidence
209Claude 3.7 Sonnet0.0%estimated ± 8.6 pp, low confidence
210Claude 3.7 Sonnet Thinking0.0%estimated ± 8.6 pp, low confidence
211Claude Opus 4.50.0%estimated ± 8.3 pp, low confidence
212Claude Sonnet 4.60.0%estimated ± 8.3 pp, low confidence
213DeepSeek V4.1 Flash 10170.0%estimated ± 8.3 pp, low confidence
214DeepSeek V4 Flash 07310.0%estimated ± 8.3 pp, low confidence
215DeepSeek V4 Pro0.0%estimated ± 8.3 pp, low confidence
216dots3-note Preview0.0%estimated ± 8.3 pp, medium confidence
217Gemini 3.1 Flash-Lite0.0%estimated ± 8.6 pp, low confidence
218Gemini 3.5 Flash0.0%estimated ± 8.3 pp, low confidence
219Gemini 3.5 Flash-Lite0.0%estimated ± 8.6 pp, low confidence
220GLM-4.50.0%estimated ± 8.6 pp, low confidence
221GLM-4.60.0%estimated ± 8.6 pp, low confidence
222GLM-4.70.0%estimated ± 8.6 pp, low confidence
223GLM-50.0%estimated ± 8.3 pp, low confidence
224GLM-5.10.0%estimated ± 8.3 pp, low confidence
225GLM-5.30.0%estimated ± 8.6 pp, low confidence
226GLM-5.3-Flash0.0%estimated ± 8.6 pp, low confidence
227GPT-4.10.0%estimated ± 8.6 pp, low confidence
228GPT-50.0%estimated ± 8.6 pp, low confidence
229GPT-5.10.0%estimated ± 8.6 pp, low confidence
230Grok 40.0%estimated ± 8.6 pp, low confidence
231Grok 4.1 Fast (Reasoning)0.0%estimated ± 8.6 pp, low confidence
232Grok 4.30.0%estimated ± 8.3 pp, low confidence
233Hy3 Preview0.0%estimated ± 8.6 pp, low confidence
234Interfaze Beta0.0%estimated ± 8.6 pp, medium confidence
235K-EXAONE 2.00.0%estimated ± 8.6 pp, low confidence
236Kimi K2.60.0%estimated ± 8.3 pp, low confidence
237Kimi K2.50.0%estimated ± 8.3 pp, low confidence
238Ling 3.0 Flash0.0%estimated ± 8.6 pp, low confidence
239Ling 3.0 Flash FP80.0%estimated ± 8.6 pp, low confidence
240Ling 3.1 Flash0.0%estimated ± 8.3 pp, low confidence
241MAI-Thinking-10.0%estimated ± 8.6 pp, low confidence
242Mercury 30.0%estimated ± 8.6 pp, medium confidence
243MiMo-V2.50.0%estimated ± 8.6 pp, low confidence
244MiniMax M2.70.0%estimated ± 8.6 pp, low confidence
245MiniMax M30.0%estimated ± 8.6 pp, low confidence
246o10.0%estimated ± 8.6 pp, low confidence
247o30.0%estimated ± 8.6 pp, low confidence
248o4-mini0.0%estimated ± 8.6 pp, low confidence
249OrcaCyber-Nano-1.00.0%estimated ± 8.6 pp, low confidence
250Pareto 26.10 Preview0.0%estimated ± 8.6 pp, medium confidence
251Particle-10.0%estimated ± 8.6 pp, medium confidence
252Qwen3.5 Flash0.0%estimated ± 8.6 pp, low confidence
253Qwen3.7 Max0.0%estimated ± 8.3 pp, low confidence
254Qwen3.7 Plus0.0%estimated ± 8.3 pp, low confidence
255Qwen3.8-Omni-Flash0.0%estimated ± 8.3 pp, low confidence
256Qwen3 Max0.0%estimated ± 8.6 pp, low confidence
257Ripple-65B Preview0.0%estimated ± 8.6 pp, low confidence
258Solar Pro 40.0%estimated ± 8.6 pp, medium confidence
259Step 5 Preview0.0%estimated ± 8.3 pp, low confidence
260Ternary Bonsai 2 27B0.0%estimated ± 8.6 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General