benchgap
Knowledge & reasoning

ARC-AGI-2 leaderboard

As of 2026-11-21, the highest measured score on ARC-AGI-2 is 95.8% by GPT-6.2 Astra. 232 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra95.8%measured
2GPT-6 Astra95.0%measured
3GPT-6.1 Sol94.2%measured
4Muse Wonder Full93.8%estimated ± 14.4 pp, low confidence
5Claude Fable 5.292.7%estimated ± 9.4 pp, medium confidence
6GPT-5.6 Sol92.5%measured
7Gemini 4 Carbon92.1%measured
8Claude Opus 5.591.7%measured
9Gemini 4 Argon91.3%measured
10Claude Opus 590.4%measured
11Gemini 4 Neon90.1%estimated ± 14.6 pp, low confidence
12Claude Fable 5.190.0%measured
13GPT-6 Sol89.6%measured
14Claude Fable 589.2%measured
15Gemini 3.8 Flash89.2%measured
16Sakana Fugu87.1%estimated ± 9.7 pp, medium confidence
17Sakana Fugu-Ultra87.1%estimated ± 9.7 pp, medium confidence
18GPT-6.2 Sol85.8%measured
19GPT-5.585.0%measured
20Muse Wonder85.0%estimated ± 14.4 pp, low confidence
21Gemini 3.7 Flash84.6%measured
22GPT-6.1 Terra84.4%estimated ± 14.4 pp, low confidence
23GPT-5.5 Pro84.2%measured
24Claude Sonnet 5.583.8%estimated ± 13.0 pp, low confidence
25Gemini 4 Hydrogen83.7%estimated ± 14.6 pp, low confidence
26GPT-5.4 Pro83.3%measured
27GPT-5.6 Terra82.5%measured
28Muse Spark 1.182.0%estimated ± 13.0 pp, low confidence
29Qwen4 Max81.7%estimated ± 13.0 pp, low confidence
30MiMo-V2.6-Pro81.7%estimated ± 14.4 pp, low confidence
31Inkling 2 Small81.7%measured
32dots3-note Preview81.4%measured
33Pareto 26.980.0%estimated ± 13.0 pp, low confidence
34Kimi K3.1 Pro79.9%estimated ± 9.7 pp, medium confidence
35Claude Mythos 579.2%estimated ± 10.9 pp, low confidence
36Muse Spark 1.378.4%estimated ± 14.4 pp, low confidence
37ERNIE 5.5 Flash Preview77.7%estimated ± 13.0 pp, low confidence
38Gemini 3.1 Pro77.1%measured
39Step 5 Preview76.1%estimated ± 9.7 pp, medium confidence
40Claude Opus 4.775.8%measured
41GPT-5.474.0%measured
42MiniMax M3.1 Flash73.9%estimated ± 13.0 pp, low confidence
43Claude Sonnet 573.6%estimated ± 13.0 pp, low confidence
44Ornith-1.5-397B72.3%estimated ± 9.7 pp, medium confidence
45Gemini 3.5 Flash72.1%measured
46Claude Opus 4.872.1%measured
47MiniMax M371.5%estimated ± 13.8 pp, low confidence
48Qwen3.8 Max71.2%estimated ± 9.7 pp, medium confidence
49Pareto 26.10 Preview70.1%estimated ± 9.7 pp, medium confidence
50Qwen3.7 Max70.1%estimated ± 9.7 pp, medium confidence
51Hy4 preview69.6%estimated ± 9.7 pp, medium confidence
52Sage-A1.569.0%estimated ± 14.4 pp, low confidence
53Claude Opus 4.6 (Adaptive)68.8%measured
54Grok 4.667.1%measured
55Qwen3.8-Flash-Next66.3%estimated ± 9.7 pp, medium confidence
56Ling 3.1 Flash65.2%estimated ± 14.4 pp, low confidence
57Muse Ember64.9%estimated ± 14.6 pp, low confidence
58Muse Spark 1.264.6%estimated ± 14.4 pp, low confidence
59Qwen3.8 Max Preview64.6%estimated ± 14.4 pp, low confidence
60GLM-5.263.5%estimated ± 9.7 pp, medium confidence
61GPT-5.3 Codex63.1%estimated ± 14.4 pp, low confidence
62Qwen3.8-Omni-Flash62.4%estimated ± 9.7 pp, medium confidence
63DeepSeek V4.1 Flash 101761.9%estimated ± 9.7 pp, medium confidence
64DeepSeek V4 Flash 073161.4%measured
65Grok 4.761.4%measured
66Claude Haiku 5.561.3%measured
67DeepSeek V4 Pro61.3%measured
68Gemini 3.6 Flash60.4%measured
69Kimi K360.4%measured
70Beam 1.0 501B-A23B60.3%estimated ± 9.7 pp, medium confidence
71Kimi K2.659.7%estimated ± 9.7 pp, medium confidence
72GPT-5.6 Luna59.5%measured
73GPT-6 Luna59.3%measured
74Muse Spark Open59.2%estimated ± 14.4 pp, low confidence
75Particle-159.2%estimated ± 9.7 pp, medium confidence
76Qwen3.7 Plus58.6%estimated ± 9.7 pp, medium confidence
77Beam 1.0 501B-A23B FP858.6%estimated ± 9.7 pp, medium confidence
78Claude Sonnet 4.658.3%measured
79Interfaze Beta56.4%estimated ± 9.7 pp, medium confidence
80Nemotron 3.5 Super VL 120B A12B NVFP456.4%estimated ± 9.7 pp, medium confidence
81Qwen3.6 Plus55.7%estimated ± 10.9 pp, low confidence
82Nemotron 3.5 Super VL 120B A12B54.4%estimated ± 9.7 pp, medium confidence
83Grok 4.353.8%estimated ± 10.9 pp, low confidence
84MiMo-V2.6-Flash53.6%estimated ± 14.4 pp, low confidence
85Grok 4.2053.3%measured
86Beam 1.0 501B-A23B NVFP453.1%estimated ± 9.7 pp, medium confidence
87GPT-5.252.9%measured
88Mercury 352.8%estimated ± 9.7 pp, medium confidence
89Grok 4.552.6%measured
90Claude Opus 4.652.6%estimated ± 9.7 pp, medium confidence
91Ornith-1.5-35B-A3B52.6%estimated ± 9.7 pp, medium confidence
92Qwen3.8-27B52.6%estimated ± 9.7 pp, medium confidence
93Solar Pro 451.5%estimated ± 9.7 pp, medium confidence
94GLM-5.351.2%estimated ± 13.8 pp, low confidence
95Grok 451.2%estimated ± 13.8 pp, low confidence
96Nemotron 3.5 Bolt 74B-A9B BF1650.6%estimated ± 9.7 pp, medium confidence
97Gemini 3 Flash50.3%estimated ± 13.8 pp, low confidence
98Kimi K2.7 Code49.8%estimated ± 14.4 pp, low confidence
99Mistral Large 449.8%estimated ± 14.4 pp, low confidence
100MiMo-V2.5-Pro49.7%estimated ± 13.0 pp, low confidence
101Quasar 438B48.6%estimated ± 14.4 pp, low confidence
102OrcaCyber-Nano-1.047.6%estimated ± 9.7 pp, medium confidence
103GPT-5.2-Codex47.3%estimated ± 14.4 pp, low confidence
104Inkling45.5%estimated ± 9.7 pp, low confidence
105o3-pro45.2%estimated ± 14.6 pp, low confidence
106Kimi K2.543.9%estimated ± 9.7 pp, low confidence
107GLM-5.3-Flash43.6%estimated ± 13.8 pp, low confidence
108Qwen3.5 397B43.0%estimated ± 10.9 pp, low confidence
109Muse Spark42.5%measured
110Hy3 Preview41.7%estimated ± 9.7 pp, low confidence
111GPT-5.1-Codex41.5%estimated ± 14.4 pp, low confidence
112GPT-5.1-Codex-Max41.5%estimated ± 14.4 pp, low confidence
113MiniMax M2.740.6%estimated ± 9.7 pp, low confidence
114Nemotron 3 Ultra40.6%estimated ± 9.7 pp, low confidence
115Inkling-Small40.1%measured
116GPT-540.1%estimated ± 13.8 pp, low confidence
117Hy340.0%estimated ± 14.4 pp, low confidence
118Claude 4.1 Opus39.7%estimated ± 14.6 pp, low confidence
119Apodex 1.139.4%estimated ± 14.4 pp, low confidence
120Apodex 1.1 Mini39.4%estimated ± 14.4 pp, low confidence
121Grok 4 Fast (Reasoning)39.2%estimated ± 13.8 pp, low confidence
122Qwen3.6-27B39.2%estimated ± 10.9 pp, low confidence
123Inkling 2 Mini38.4%estimated ± 9.7 pp, low confidence
124Qwen 3.6 Max (preview)37.7%estimated ± 14.4 pp, low confidence
125Claude Opus 4.5 Thinking37.6%measured
126Nemotron 3.5 Bolt 74B-A9B37.5%estimated ± 9.7 pp, low confidence
127Ornith-1.5-9B37.3%estimated ± 9.7 pp, low confidence
128Solar Open 236.7%estimated ± 9.7 pp, low confidence
129GLM-5.136.2%estimated ± 9.7 pp, low confidence
130Muse Glimmer 30B35.6%estimated ± 14.4 pp, low confidence
131GLM-535.1%estimated ± 9.7 pp, low confidence
132Step 3.7 Flash35.0%estimated ± 14.4 pp, low confidence
133Ling 3.0 Flash VL34.4%estimated ± 14.4 pp, low confidence
134Grok 4.1 Fast (Reasoning)34.3%estimated ± 13.8 pp, low confidence
135Claude Opus 4.534.1%estimated ± 10.9 pp, low confidence
136Ternary Bonsai 2 27B33.8%estimated ± 9.7 pp, low confidence
137o333.5%estimated ± 13.8 pp, low confidence
138Gemini 2.5 Flash33.2%estimated ± 14.4 pp, low confidence
139A.X K232.9%estimated ± 9.7 pp, low confidence
140Claude 4 Sonnet32.7%estimated ± 14.4 pp, low confidence
141K-Exaone32.7%estimated ± 14.4 pp, low confidence
142MiMo-V2-Omni32.7%estimated ± 14.4 pp, low confidence
143DeepSeek V3.232.3%estimated ± 14.4 pp, low confidence
144Trinity-Large-Preview32.3%estimated ± 14.4 pp, low confidence
145Trinity-Large-Thinking32.3%estimated ± 14.4 pp, low confidence
146Gemini 3.5 Flash-Lite32.1%estimated ± 13.8 pp, low confidence
147Gemma 4 E4B31.7%estimated ± 14.4 pp, low confidence
148GLM-5V-Turbo31.7%estimated ± 14.4 pp, low confidence
149Qwen3.5-122B-A10B31.5%estimated ± 10.9 pp, low confidence
150Command A+31.1%estimated ± 14.4 pp, low confidence
151GLM-5-Turbo31.1%estimated ± 14.4 pp, low confidence
152Granite 4.2 30B31.1%estimated ± 14.4 pp, low confidence
153Granite 4.2 8B31.1%estimated ± 14.4 pp, low confidence
154MiMo-V2-Pro31.1%estimated ± 14.4 pp, low confidence
155MiniCPM5-2B31.1%estimated ± 14.4 pp, low confidence
156Mistral Small 431.1%estimated ± 14.4 pp, low confidence
157North Mini Code31.1%estimated ± 14.4 pp, low confidence
158Sarvam 30B31.1%estimated ± 14.4 pp, low confidence
159Gemini 3 Pro31.1%measured
160Ripple-65B Preview30.7%estimated ± 9.7 pp, low confidence
161Celeris-130.5%estimated ± 14.4 pp, low confidence
162Claude 3.5 Haiku30.5%estimated ± 14.4 pp, low confidence
163Claude 3 Haiku30.5%estimated ± 14.4 pp, low confidence
164DeepSeek V330.5%estimated ± 14.4 pp, low confidence
165DeepSeek V3 032430.5%estimated ± 14.4 pp, low confidence
166Devstral Small30.5%estimated ± 14.4 pp, low confidence
167Exaone 4.0 1.2B30.5%estimated ± 14.4 pp, low confidence
168Exaone 4.0 32B30.5%estimated ± 14.4 pp, low confidence
169Gemini 2.5 Flash-Lite30.5%estimated ± 14.4 pp, low confidence
170Gemma 3 27B30.5%estimated ± 14.4 pp, low confidence
171Gemma 4 E2B30.5%estimated ± 14.4 pp, low confidence
172GLM-4.5-Air30.5%estimated ± 14.4 pp, low confidence
173GPT-4.1 nano30.5%estimated ± 14.4 pp, low confidence
174GPT-4o30.5%estimated ± 14.4 pp, low confidence
175Granite-4.0-350M30.5%estimated ± 14.4 pp, low confidence
176Granite-4.0-H-1B30.5%estimated ± 14.4 pp, low confidence
177Granite-4.0-H-350M30.5%estimated ± 14.4 pp, low confidence
178Granite 4.2 3B30.5%estimated ± 14.4 pp, low confidence
179Grok 4.1 Fast30.5%estimated ± 14.4 pp, low confidence
180Grok Code Fast 130.5%estimated ± 14.4 pp, low confidence
181LFM2.5-2.6B30.5%estimated ± 14.4 pp, low confidence
182LFM2.5-8B-A1B30.5%estimated ± 14.4 pp, low confidence
183LFM2.5-VL-1.6B-Extract30.5%estimated ± 14.4 pp, low confidence
184Ling 2.6 Flash30.5%estimated ± 14.4 pp, low confidence
185Ling 3.0 Tiny30.5%estimated ± 14.4 pp, low confidence
186Llama 3.1 405B30.5%estimated ± 14.4 pp, low confidence
187Llama 3.1 70B30.5%estimated ± 14.4 pp, low confidence
188Llama 3.3 70B30.5%estimated ± 14.4 pp, low confidence
189Llama 4 Maverick30.5%estimated ± 14.4 pp, low confidence
190Llama 4 Scout30.5%estimated ± 14.4 pp, low confidence
191Magistral Small30.5%estimated ± 14.4 pp, low confidence
192Mercury 2.530.5%estimated ± 14.4 pp, low confidence
193Mistral Large 230.5%estimated ± 14.4 pp, low confidence
194Mistral Medium 330.5%estimated ± 14.4 pp, low confidence
195Mistral Medium 3.5 128B30.5%estimated ± 14.4 pp, low confidence
196Mistral Small 3.130.5%estimated ± 14.4 pp, low confidence
197Nova Pro30.5%estimated ± 14.4 pp, low confidence
198Phi-430.5%estimated ± 14.4 pp, low confidence
199Phi-4-reasoning30.5%estimated ± 14.4 pp, low confidence
200Qwen3.5-4B30.5%estimated ± 14.4 pp, low confidence
201Qwen3-Omni-30B-A3B-Instruct30.5%estimated ± 14.4 pp, low confidence
202Qwen3-Omni-30B-A3B-Thinking30.5%estimated ± 14.4 pp, low confidence
203Sarvam 105B30.5%estimated ± 14.4 pp, low confidence
204Solar Pro 230.5%estimated ± 14.4 pp, low confidence
205Solar Pro 330.5%estimated ± 14.4 pp, low confidence
206Ultravox v0.6 Llama 3.3 70B30.5%estimated ± 14.4 pp, low confidence
207Ling 3.0 Flash29.5%estimated ± 9.7 pp, low confidence
208GPT-5.128.3%estimated ± 5.2 pp, medium confidence
209Qwen3.6-35B-A3B27.7%estimated ± 10.9 pp, low confidence
210Qwen3.5 Flash27.7%estimated ± 13.8 pp, low confidence
211o1-preview26.4%estimated ± 14.6 pp, low confidence
212QwQ-32B26.1%estimated ± 14.4 pp, low confidence
213Phi-4 Multimodal Instruct26.1%estimated ± 14.4 pp, low confidence
214DeepSeek R1 Distill Qwen 32B26.1%estimated ± 14.4 pp, low confidence
215Gemini 1.5 Pro26.1%estimated ± 14.4 pp, low confidence
216Llama 3.1 8B26.1%estimated ± 14.4 pp, low confidence
217Gemini 1.0 Pro26.1%estimated ± 14.4 pp, low confidence
218GPT-4o mini26.1%estimated ± 14.4 pp, low confidence
219o1-mini26.1%estimated ± 14.4 pp, low confidence
220Qwen2.5 Coder 32B Instruct26.1%estimated ± 14.4 pp, low confidence
221Gemini 1.5 Flash26.1%estimated ± 14.4 pp, low confidence
222GPT-4 Turbo26.1%estimated ± 14.4 pp, low confidence
223Claude 3 Opus26.1%estimated ± 14.4 pp, low confidence
224GLM-4.725.8%estimated ± 10.9 pp, low confidence
225MAI-Thinking-125.3%estimated ± 9.7 pp, low confidence
226Qwen3.5-27B24.5%estimated ± 10.9 pp, low confidence
227Ling 3.0 Flash FP824.2%estimated ± 9.7 pp, low confidence
228MiMo-V2.522.4%estimated ± 13.8 pp, low confidence
229Gemini 3.1 Flash-Lite20.2%estimated ± 13.8 pp, low confidence
230GPT-5.4 mini18.9%measured
231Gemma 4 31B16.9%estimated ± 10.9 pp, low confidence
232Qwen3.5-35B-A3B16.3%estimated ± 10.9 pp, low confidence
233DeepSeek-V2.516.2%estimated ± 14.6 pp, low confidence
234K-EXAONE 2.014.3%estimated ± 9.7 pp, low confidence
235Claude Sonnet 4.5 Thinking13.6%measured
236Claude Sonnet 4.513.6%measured
237Qwen3 Max13.1%estimated ± 13.8 pp, low confidence
238MiMo-V2-Flash13.1%estimated ± 10.9 pp, low confidence
239Gemini 2.5 Pro8.6%estimated ± 10.9 pp, low confidence
240GPT-5.4 nano5.7%measured
241Gemma 4 26B A4B4.2%estimated ± 10.9 pp, low confidence
242Claude 3.7 Sonnet0.0%estimated ± 13.8 pp, low confidence
243Claude 3.7 Sonnet Thinking0.0%estimated ± 10.9 pp, low confidence
244Claude 4 Opus0.0%estimated ± 10.9 pp, low confidence
245Claude 4 Opus Thinking0.0%estimated ± 10.9 pp, low confidence
246Claude Haiku 4.50.0%estimated ± 13.8 pp, low confidence
247DeepSeek-R10.0%estimated ± 10.9 pp, low confidence
248DeepSeek-R1-05280.0%estimated ± 10.9 pp, low confidence
249DeepSeek V3.10.0%estimated ± 10.9 pp, low confidence
250Gemini 2.0 Flash0.0%estimated ± 13.8 pp, low confidence
251Gemma 4 12B0.0%estimated ± 10.9 pp, low confidence
252GLM-4.50.0%estimated ± 13.8 pp, low confidence
253GLM-4.60.0%estimated ± 13.8 pp, low confidence
254GLM-4.7-Flash0.0%estimated ± 10.9 pp, low confidence
255GPT-4.10.0%estimated ± 13.8 pp, low confidence
256GPT-4.1 mini0.0%estimated ± 13.8 pp, low confidence
257GPT-4.50.0%estimated ± 10.9 pp, low confidence
258GPT-OSS 120B0.0%estimated ± 10.9 pp, low confidence
259GPT-OSS 20B0.0%estimated ± 10.9 pp, low confidence
260Grok 3 [Beta]0.0%estimated ± 10.9 pp, low confidence
261Grok 3 Mini0.0%estimated ± 10.9 pp, low confidence
262Kimi K20.0%estimated ± 10.9 pp, low confidence
263LFM2.5-24B-A2B0.0%estimated ± 10.9 pp, low confidence
264Magistral Medium0.0%estimated ± 10.9 pp, low confidence
265Mistral Large 30.0%estimated ± 13.8 pp, low confidence
266Nemotron 3.5 Lightning 30B A3B NVFP40.0%estimated ± 10.9 pp, low confidence
267Nemotron 3 Nano 30B0.0%estimated ± 10.9 pp, low confidence
268Nemotron 3 Nano Omni 30B A3B0.0%estimated ± 10.9 pp, low confidence
269Nemotron 3 Super 100B0.0%estimated ± 10.9 pp, low confidence
270Nemotron Ultra 253B0.0%estimated ± 10.9 pp, low confidence
271o10.0%estimated ± 10.9 pp, low confidence
272o1-pro0.0%estimated ± 10.9 pp, low confidence
273o3-mini0.0%estimated ± 10.9 pp, low confidence
274o4-mini0.0%estimated ± 10.9 pp, low confidence
275Phi-4-reasoning-plus0.0%estimated ± 10.9 pp, low confidence
276Qwen3 235B 25070.0%estimated ± 10.9 pp, low confidence
277Qwen3 235B-A22B (Reasoning)0.0%estimated ± 10.9 pp, low confidence
278Spark-X2.6-4B0.0%estimated ± 10.9 pp, low confidence
279ZAYA1-8B0.0%estimated ± 10.9 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General