benchgap
Knowledge & reasoning

GPQA leaderboard

As of 2026-11-21, the highest measured score on GPQA is 97.0% by GPT-6.2 Astra. 172 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra97.0%measured
2Claude Fable 5.296.5%estimated ± 0.9 pp, high confidence
3Gemini 4 Carbon96.5%estimated ± 0.9 pp, high confidence
4Claude Sonnet 5.596.4%estimated ± 1.2 pp, low confidence
5Claude Opus 5.596.3%estimated ± 1.2 pp, medium confidence
6Ling 3.1 Flash96.3%estimated ± 1.2 pp, medium confidence
7GPT-6.2 Sol96.0%measured
8GPT-6 Astra96.0%measured
9GPT-6.1 Terra95.9%estimated ± 1.2 pp, medium confidence
10GPT-6.1 Sol95.8%estimated ± 0.3 pp, medium confidence
11Muse Wonder Full95.8%estimated ± 1.2 pp, medium confidence
12Gemini 4 Argon95.5%estimated ± 0.9 pp, high confidence
13Sakana Fugu95.5%measured
14Sakana Fugu-Ultra95.5%measured
15Claude Fable 5.195.2%estimated ± 1.6 pp, high confidence
16Claude Fable 595.0%estimated ± 1.6 pp, high confidence
17Claude Opus 594.7%estimated ± 0.9 pp, high confidence
18GPT-5.6 Sol94.6%measured
19Muse Wonder94.6%estimated ± 1.2 pp, medium confidence
20Gemini 3.7 Flash94.4%estimated ± 1.6 pp, high confidence
21GPT-5.5 Pro94.4%estimated ± 1.6 pp, high confidence
22Gemini 3.1 Pro94.3%estimated ± 0.0 pp, high confidence
23GPT-5.4 Pro94.3%estimated ± 1.6 pp, high confidence
24Claude Opus 4.794.2%measured
25Kimi K3.1 Pro94.2%measured
26Qwen4 Max94.1%estimated ± 1.2 pp, medium confidence
27Claude Mythos 594.1%measured
28dots3-note Preview94.0%estimated ± 1.6 pp, high confidence
29MiniMax M393.9%estimated ± 4.5 pp, high confidence
30Pareto 26.993.8%estimated ± 3.2 pp, high confidence
31Muse Spark 1.193.6%estimated ± 1.2 pp, medium confidence
32GPT-6 Luna93.6%estimated ± 0.3 pp, medium confidence
33Muse Spark 1.393.6%estimated ± 6.1 pp, medium confidence
34Gemini 3.8 Flash93.6%estimated ± 0.9 pp, high confidence
35Claude Opus 4.893.6%measured
36GPT-5.593.6%measured
37ERNIE 5.5 Flash Preview93.6%estimated ± 3.2 pp, high confidence
38Step 5 Preview93.5%estimated ± 0.0 pp, high confidence
39Kimi K393.5%measured
40GPT-6 Sol93.3%estimated ± 0.9 pp, high confidence
41MiniMax M3.1 Flash93.2%estimated ± 3.2 pp, high confidence
42Grok 4.693.2%estimated ± 0.9 pp, high confidence
43Grok 4.793.2%estimated ± 0.9 pp, high confidence
44Claude Haiku 5.593.1%estimated ± 0.9 pp, high confidence
45Grok 4.593.1%estimated ± 0.9 pp, high confidence
46Gemini 4 Neon93.1%estimated ± 7.8 pp, medium confidence
47Inkling 2 Small93.0%measured
48Qwen3.8 Max Preview93.0%estimated ± 6.1 pp, medium confidence
49Gemini 4 Hydrogen93.0%estimated ± 7.8 pp, medium confidence
50GPT-5.6 Terra92.9%measured
51GPT-5.492.8%measured
52Ornith-1.5-397B92.8%measured
53MiMo-V2.6-Pro92.8%estimated ± 7.8 pp, medium confidence
54Qwen3.8 Max92.6%measured
55Ling 3.0 Flash Sante92.6%estimated ± 1.2 pp, medium confidence
56Sage-A1.592.4%estimated ± 11.9 pp, low confidence
57Pareto 26.10 Preview92.4%estimated ± 0.0 pp, high confidence
58GPT-5.292.4%measured
59Qwen3.7 Max92.4%measured
60Muse Spark 1.292.4%estimated ± 5.1 pp, medium confidence
61Claude Opus 4.6 (Adaptive)92.3%estimated ± 1.6 pp, high confidence
62Hy492.3%estimated ± 1.9 pp, high confidence
63GPT-5.6 Luna92.3%measured
64Hy4 preview92.3%measured
65Claude Sonnet 592.2%estimated ± 1.9 pp, high confidence
66Gemini 3.5 Flash92.2%measured
67Apodex 1.192.2%estimated ± 1.9 pp, high confidence
68Hy4 flash92.1%estimated ± 1.9 pp, high confidence
69Muse Ember92.1%estimated ± 7.8 pp, medium confidence
70GPT-5.3 Codex91.8%estimated ± 6.1 pp, medium confidence
71Qwen3.8-Flash-Next91.7%measured
72MiMo-V2.5-Pro91.7%estimated ± 1.9 pp, high confidence
73Agents-A191.7%estimated ± 1.9 pp, high confidence
74Claude Opus 4.691.3%measured
75Gemini 3.6 Flash91.2%estimated ± 1.6 pp, high confidence
76GLM-5.291.2%measured
77Qwen3.8-Omni-Flash91.0%measured
78DeepSeek V4.1 Flash 101790.9%measured
79Beam 1.0 501B-A23B90.6%estimated ± 0.0 pp, medium confidence
80Kimi K2.690.5%measured
81Particle-190.4%estimated ± 0.0 pp, medium confidence
82Qwen3.6 Plus90.4%measured
83GPT-5.2-Codex90.4%estimated ± 6.1 pp, medium confidence
84Beam 1.0 501B-A23B FP890.3%estimated ± 0.0 pp, medium confidence
85Qwen3.7 Plus90.3%measured
86Hy390.2%estimated ± 6.1 pp, medium confidence
87DeepSeek V4 Pro90.1%measured
88Grok 4.390.1%measured
89Kimi K2.7 Code90.1%estimated ± 6.1 pp, medium confidence
90Nemotron 3.5 Super VL 120B A12B NVFP489.9%estimated ± 0.0 pp, medium confidence
91Claude Sonnet 4.689.9%measured
92Interfaze Beta89.9%measured
93MiMo-V2.6-Flash89.8%estimated ± 7.8 pp, medium confidence
94GLM-5.389.7%estimated ± 4.5 pp, high confidence
95Grok 489.7%estimated ± 4.5 pp, high confidence
96Mistral Large 489.6%estimated ± 7.8 pp, medium confidence
97Nemotron 3.5 Super VL 120B A12B89.5%estimated ± 0.0 pp, medium confidence
98Muse Spark Open89.5%estimated ± 6.1 pp, medium confidence
99Qwen 3.6 Max (preview)89.5%estimated ± 2.4 pp, high confidence
100Muse Spark89.5%estimated ± 0.0 pp, medium confidence
101Inkling-Small89.5%measured
102Gemini 3 Flash89.5%estimated ± 4.5 pp, high confidence
103Beam 1.0 501B-A23B NVFP489.3%estimated ± 0.0 pp, medium confidence
104Mercury 389.3%estimated ± 0.0 pp, medium confidence
105Ornith-1.5-35B-A3B89.2%measured
106Qwen3.8-27B89.2%measured
107Solar Pro 489.0%estimated ± 0.0 pp, medium confidence
108Nemotron 3.5 Bolt 74B-A9B BF1688.9%estimated ± 0.0 pp, medium confidence
109Grok 4.2088.5%estimated ± 0.0 pp, medium confidence
110Qwen3.5 397B88.4%measured
111OrcaCyber-Nano-1.088.3%estimated ± 0.0 pp, medium confidence
112Seed 2.1 Pro88.3%estimated ± 2.4 pp, high confidence
113Claude Opus 4.5 Thinking88.2%estimated ± 1.6 pp, high confidence
114GLM-5.3-Flash88.1%estimated ± 4.5 pp, high confidence
115DeepSeek V4 Flash 073188.1%measured
116GPT-5.188.1%estimated ± 1.8 pp, high confidence
117GPT-5.4 mini88.0%measured
118Inkling87.9%measured
119Qwen3.6-27B87.8%measured
120MiMo-V2-Pro87.7%estimated ± 6.1 pp, medium confidence
121Kimi K2.587.6%measured
122GPT-587.4%estimated ± 4.5 pp, high confidence
123Gemini 3 Pro87.3%estimated ± 1.6 pp, high confidence
124Hy3 Preview87.2%measured
125Apodex 1.1 Mini87.2%estimated ± 6.1 pp, medium confidence
126Grok 4 Fast (Reasoning)87.2%estimated ± 4.5 pp, high confidence
127Ling 3.0 Flash VL87.0%estimated ± 6.1 pp, medium confidence
128Claude Opus 4.587.0%measured
129Nemotron 3 Ultra87.0%measured
130MiniMax M2.786.9%estimated ± 1.2 pp, high confidence
131GPT-5.1-Codex86.8%estimated ± 6.1 pp, medium confidence
132GPT-5.1-Codex-Max86.8%estimated ± 6.1 pp, medium confidence
133Inkling 2 Mini86.6%measured
134Qwen3.5-122B-A10B86.6%measured
135Seed 2.1 Turbo86.4%estimated ± 2.4 pp, high confidence
136Ornith-1.5-9B86.4%measured
137Nemotron 3.5 Bolt 74B-A9B86.3%estimated ± 1.2 pp, high confidence
138Solar Open 286.2%estimated ± 1.2 pp, high confidence
139Grok 4.1 Fast (Reasoning)86.2%estimated ± 4.5 pp, high confidence
140GLM-5.186.1%estimated ± 1.2 pp, high confidence
141GLM-586.0%measured
142Qwen3.6-35B-A3B86.0%measured
143o386.0%estimated ± 4.5 pp, high confidence
144Ternary Bonsai 2 27B85.8%measured
145Gemini 3.5 Flash-Lite85.7%estimated ± 4.5 pp, high confidence
146GLM-4.785.7%measured
147GLM-5-Turbo85.7%estimated ± 6.1 pp, medium confidence
148A.X K285.5%estimated ± 1.2 pp, high confidence
149Qwen3.5-27B85.5%measured
150o3-pro85.5%estimated ± 6.1 pp, medium confidence
151Ripple-65B Preview85.1%estimated ± 1.2 pp, high confidence
152Claude Sonnet 4.5 Thinking85.0%estimated ± 1.6 pp, high confidence
153Ling 3.0 Flash85.0%measured
154Qwen3.5 Flash84.8%estimated ± 4.5 pp, high confidence
155Muse Glimmer 30B84.6%estimated ± 6.1 pp, medium confidence
156Gemma 4 31B84.3%measured
157MAI-Thinking-184.2%measured
158Qwen3.5-35B-A3B84.2%measured
159Ling 3.0 Flash FP884.0%measured
160MiMo-V2-Omni83.9%estimated ± 6.1 pp, medium confidence
161MiMo-V2.583.7%estimated ± 4.5 pp, high confidence
162MiMo-V2-Flash83.7%measured
163Claude Sonnet 4.583.4%measured
164Gemini 3.1 Flash-Lite83.3%estimated ± 4.5 pp, high confidence
165Gemini 2.5 Pro83.0%measured
166Claude 4.1 Opus82.9%estimated ± 7.8 pp, medium confidence
167GPT-5.4 nano82.8%measured
168Gemma 4 26B A4B82.3%measured
169GLM-5V-Turbo82.2%estimated ± 6.1 pp, medium confidence
170Step 3.7 Flash82.2%estimated ± 6.1 pp, medium confidence
171K-EXAONE 2.082.1%estimated ± 1.2 pp, high confidence
172Qwen3 Max81.8%estimated ± 4.5 pp, high confidence
173DeepSeek-R1-052881.0%measured
174Grok 3 Mini80.3%measured
175GPT-OSS 120B80.1%measured
176K-Exaone79.9%estimated ± 6.1 pp, medium confidence
177Claude 4 Opus Thinking79.6%measured
178Nemotron 3 Super 100B79.2%measured
179o1-pro79.0%measured
180Mercury 2.578.9%estimated ± 1.2 pp, high confidence
181Gemma 4 12B78.8%measured
182Qwen3.5-4B78.8%estimated ± 6.1 pp, medium confidence
183Mistral Small 478.6%estimated ± 6.1 pp, medium confidence
184Sarvam 105B78.6%estimated ± 1.2 pp, high confidence
185o1-preview78.2%estimated ± 6.1 pp, medium confidence
186Claude 3.7 Sonnet Thinking78.2%measured
187Command A+77.9%estimated ± 6.1 pp, medium confidence
188Exaone 4.0 32B77.7%estimated ± 5.7 pp, medium confidence
189o4-mini77.6%measured
190North Mini Code77.5%estimated ± 6.1 pp, medium confidence
191Qwen3 235B 250777.5%measured
192GLM-4.677.2%estimated ± 4.5 pp, high confidence
193o3-mini77.2%measured
194DeepSeek V3.277.0%estimated ± 6.1 pp, medium confidence
195Trinity-Large-Thinking76.2%estimated ± 1.2 pp, high confidence
196Nemotron Ultra 253B76.0%measured
197o175.7%measured
198Nemotron 3.5 Lightning 30B A3B NVFP475.6%measured
199Ling 3.0 Tiny75.4%estimated ± 6.1 pp, medium confidence
200Grok 3 [Beta]75.4%measured
201GLM-4.5-Air75.3%estimated ± 6.1 pp, medium confidence
202Quasar 438B75.3%estimated ± 6.1 pp, medium confidence
203GLM-4.7-Flash75.2%measured
204Kimi K275.1%measured
205Claude Haiku 4.575.1%estimated ± 4.5 pp, high confidence
206GLM-4.575.1%estimated ± 4.5 pp, high confidence
207Claude 4 Opus74.9%measured
208DeepSeek V3.174.9%measured
209Grok Code Fast 174.8%estimated ± 6.1 pp, medium confidence
210Qwen3-Omni-30B-A3B-Thinking74.7%estimated ± 6.1 pp, medium confidence
211Solar Pro 374.5%estimated ± 6.1 pp, medium confidence
212LFM2.5-24B-A2B74.4%measured
213Nemotron 3 Nano 30B73.0%measured
214Nemotron 3 Nano Omni 30B A3B72.2%measured
215Mistral Large 371.6%estimated ± 4.5 pp, high confidence
216DeepSeek-R171.5%measured
217GPT-OSS 20B71.5%measured
218Qwen3 235B-A22B (Reasoning)71.1%measured
219ZAYA1-8B71.0%measured
220Claude 4 Sonnet70.8%estimated ± 6.1 pp, medium confidence
221Gemini 2.5 Flash70.8%estimated ± 6.1 pp, medium confidence
222Magistral Medium70.8%measured
223MiniCPM5-2B70.0%estimated ± 1.2 pp, high confidence
224Spark-X2.6-4B69.9%measured
225Llama 4 Maverick69.7%estimated ± 6.1 pp, medium confidence
226GPT-4.569.5%measured
227LongCat-Flash-Lite-Sparse69.3%estimated ± 1.2 pp, high confidence
228Phi-4-reasoning-plus68.9%measured
229Qwen3 32B (Reasoning)68.4%measured
230DeepSeek V3 032468.3%estimated ± 6.1 pp, medium confidence
231Magistral Small68.2%measured
232Claude 3.7 Sonnet68.0%measured
233Celeris-167.3%estimated ± 5.7 pp, medium confidence
234Grok 4.1 Fast66.7%estimated ± 6.1 pp, medium confidence
235Granite 4.2 30B66.4%measured
236Sarvam 30B66.3%estimated ± 1.2 pp, high confidence
237GPT-4.166.3%measured
238Phi-4-reasoning65.8%measured
239QwQ-32B65.6%measured
240DeepSeek R1 Distill Llama 70B65.2%measured
241Gemini 2.0 Flash65.2%measured
242Qwen3-Omni-30B-A3B-Instruct65.1%estimated ± 6.1 pp, medium confidence
243DeepSeek R1 Distill Qwen 32B64.7%estimated ± 6.1 pp, medium confidence
244Gemini 2.5 Flash-Lite64.6%measured
245Mellum2.1-12B-A2.5B-Thinking64.4%estimated ± 1.2 pp, high confidence
246GPT-4.1 mini64.2%measured
247Granite 4.2 8B64.1%measured
248GPT-4 Turbo63.9%estimated ± 7.8 pp, medium confidence
249LFM3-1.7B63.4%measured
250Trinity-Large-Preview63.1%estimated ± 1.2 pp, high confidence
251Qwen3 235B-A22B62.9%measured
252Gemini 1.5 Pro62.4%estimated ± 4.5 pp, high confidence
253Llama 4 Scout62.1%estimated ± 6.1 pp, medium confidence
254Mistral Medium 361.3%estimated ± 6.1 pp, medium confidence
255DeepSeek-R1-0528-Qwen3-8B61.1%measured
256Qwen2.5-Max60.1%measured
257o1-mini60.0%measured
258Solar Pro 259.8%estimated ± 6.1 pp, medium confidence
259LFM2.5-2.6B59.5%estimated ± 6.1 pp, medium confidence
260Laguna XS.259.4%estimated ± 4.5 pp, high confidence
261Claude 3.5 Sonnet59.4%measured
262DeepSeek R1 Distill Qwen 14B59.1%measured
263DeepSeek V359.1%measured
264Ling 2.6 Flash59.0%measured
265Gemma 4 E4B58.6%measured
266GPT-4o58.2%estimated ± 4.5 pp, high confidence
267Mellum2-12B-A2.5B-Thinking57.6%measured
268ZAYA1-74B-Preview57.3%measured
269Phi-456.1%measured
270Grok 256.0%measured
271Llama 3.1 405B55.6%estimated ± 6.1 pp, medium confidence
272LFM2.5-8B-A1B55.5%estimated ± 6.1 pp, medium confidence
273Granite 4.2 3B54.8%measured
274Qwen3 32B54.6%measured
275MiniMax-Text-0154.4%measured
276Nova Pro54.2%estimated ± 6.1 pp, medium confidence
277Ultravox v0.6 Llama 3.3 70B54.1%estimated ± 6.1 pp, medium confidence
278Claude 3 Opus53.3%estimated ± 6.1 pp, medium confidence
279Mistral Large 253.0%estimated ± 6.1 pp, medium confidence
280Grok 2 mini51.0%measured
281Gemini 2.0 Flash-Lite50.5%measured
282Llama 3.3 70B50.5%measured
283GPT-4.1 nano50.3%measured
284Gemini 1.5 Flash50.0%measured
285GPT-4o mini49.4%estimated ± 4.5 pp, high confidence
286Gemma 3 27B47.8%estimated ± 6.1 pp, medium confidence
287Exaone 4.0 1.2B47.4%estimated ± 6.1 pp, medium confidence
288DeepSeek-V2.547.3%estimated ± 6.1 pp, medium confidence
289Qwen2.5 Coder 32B Instruct46.8%estimated ± 6.1 pp, medium confidence
290Llama 3.1 70B46.7%measured
291Devstral Small46.5%estimated ± 6.1 pp, medium confidence
292Qwen2.5-VL-32B46.0%measured
293Mistral Small 3.146.0%measured
294LLaDA2.2-mini44.1%estimated ± 1.2 pp, high confidence
295Gemma 4 E2B43.4%measured
296Soofi S 30B-A3B43.4%measured
297Claude 3 Haiku42.9%estimated ± 6.1 pp, medium confidence
298Claude 3.5 Haiku41.6%measured
299Mellum2-12B-A2.5B-Instruct40.9%measured
300Mistral Medium 3.5 128B40.8%estimated ± 4.5 pp, medium confidence
301Mixtral 8x22B Instruct v0.140.5%estimated ± 7.6 pp, medium confidence
302Gemma 2 27B38.1%estimated ± 7.6 pp, medium confidence
303Phi-4 Multimodal Instruct37.6%estimated ± 6.1 pp, medium confidence
304LFM2.5-VL-1.6B-Extract35.2%estimated ± 6.1 pp, medium confidence
305Gemini 1.0 Pro34.1%estimated ± 6.1 pp, medium confidence
306Laguna M.133.7%estimated ± 4.5 pp, medium confidence
307LFM2.5-350M30.6%measured
308Llama 3.1 8B30.4%measured
309Phi-3.5-mini30.4%measured
310Granite-4.0-1B29.9%measured
311Granite-4.0-H-1B29.7%measured
312Granite-4.0-H-350M26.1%measured
313MiniCPM5-1B25.9%estimated ± 1.2 pp, high confidence
314LFM2.5-VL-450M25.7%measured
315LFM2.5-230M25.4%measured
316Granite-4.0-350M24.1%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General