Inference
EN
Epoch AI

GPQA diamond

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.217
Top resultBest result on this benchmark.96.5%Gemini 4 Carbon
Top-3 spreadPoints from first to third.0.8 pts
YearYear of release.—

Result and price

122 of 217 results

Results

217 results
#ModelResultThe score from the source.
01Gemini 4 Carbonhigh effort96.5% ±1.2
02Gemini 4 Argonhigh effort96.4% ±1.2
03GPT-6 Astramax effort95.8% ±1.4
04Claude Sonnet 5.5max effort95.6% ±1.4
05GPT-6.1 Solmax effort95.4% ±1.4
05Gemini 3.8 Flashhigh effort95.4% ±1.4
07Gemini 3.7 Flashhigh effort94.8% ±1.3
08GPT-5.4 Proxhigh effort94.6% ±1.6
09Gemini 3.1 Prohigh effort94.4% ±1.6
10GPT-6.1 Terramax effort94.3% ±1.5
10GPT-6 Solmax effort94.3% ±1.5
12Gemini 3.6 Flashhigh effort94.1% ±1.4
13GPT-5.5xhigh effort94.0% ±1.5
14GPT-5.5 Proxhigh effort93.9% ±1.6
14Claude Fable 5.2max effort93.9% ±1.6
14Claude Opus 5max effort93.9% ±1.5
17GPT-5.6 Solmax effort93.5% ±1.6
18Grok 4.5high effort93.4% ±1.4
19GPT-5.6 Terramax effort93.3% ±1.5
19GPT-5.4xhigh effort93.3% ±1.8
21Grok 4.6xhigh effort93.2% ±1.5
22Kimi K3max effort93.1% ±1.5
22Doubao Seed Evolving (Nov 2026)high effort93.1% ±1.6
24Gemini 3.5 Flashhigh effort92.8% ±1.6
25Grok 4.7xhigh effort92.7% ±1.6
25Qwen3.8 Maxxhigh effort92.7% ±1.7
27Gemini 3 Pro92.6% ±1.7
28Qwen3.8 Max (0902)xhigh effort92.3% ±1.7
29GLM-5.2max effort91.9% ±1.6
30DeepSeek V4 Pro 0813max effort91.7% ±1.5
31GPT-5.6 Lunamax effort91.6% ±1.7
32GPT-5.2xhigh effort91.4% ±1.8
33Claude Opus 4.8max effort91.0% ±1.9
33DeepSeek V4 Flash 0731max effort91.0% ±1.8
35GLM-5.3max effort90.9% ±1.6
35Qwen3.7 Max90.9% ±2.0
37Kimi K2.690.8% ±1.7
38Claude Opus 5.5max effort90.6% ±1.9
39GPT-6 Lunamax effort90.5% ±1.9
40GLM-5.3-Flashmax effort90.2% ±1.7
41GLM-5.189.9% ±2.1
42DeepSeek V4.1 Flash 1017max effort89.8% ±1.7
42Muse Spark89.8% ±2.2
44Claude Haiku 5.5max effort89.6% ±2.0
44DeepSeek V4 Pro 0423max effort89.6% ±1.7
44Doubao Seed 2.1 Turbohigh effort89.6% ±1.8
47Gemini 3 Flashhigh effort89.4% ±2.2
48Grok 4.2089.3% ±1.9
48Doubao Seed 2.1 Prohigh effort89.3% ±1.8
50Mistral Large 4high effort89.1% ±1.8
51Grok 4.3 Betahigh effort88.8% ±2.0
52Muse Spark Openxhigh effort88.6% ±1.9
53Inkling-Smallxhigh effort88.5% ±1.9
54Claude Opus 4.6max effort88.4% ±2.3
54Qwen3.6 Plus88.4% ±2.3
56Inklingxhigh effort88.3% ±1.9
57Doubao-Seed-2.0 Prohigh effort88.2% ±1.9
58Doubao Seed 2.0 Lite (Apr 2026)high effort88.1% ±1.9
59Kimi K2.7 Code87.9% ±2.3
60GLM-587.8% ±2.3
61GPT-5.1high effort87.6% ±1.9
61Kimi K2.587.6% ±1.9
63Qwen 3.6 Max (preview)87.4% ±2.4
64Grok 487.0% ±2.0
65GPT-5.4 minixhigh effort86.9% ±2.4
66Claude Opus 4.7max effort86.4% ±2.4
66Qwen3.5 397B A17Bnone effort86.4% ±2.4
68GPT-5high effort86.2% ±2.1
69Claude Fable 5max effort85.9% ±2.5
70Claude Opus 4.585.5% ±2.2
71Gemini 2.5 Pro (Jun 2025)85.1% ±2.1
72Qwen3.6-27Bnone effort84.8% ±2.6
72Qwen3.6-35B-A3Bnone effort84.8% ±2.6
72Qwen 3.5 Plus (hosted 397B-A17B)84.8% ±2.6
75Kimi K2 Thinking84.2% ±2.1
76Gemini 2.5 Pro (Mar 2025)83.8% ±2.6
77Gemini 3.5 Flash-Litehigh effort83.3% ±2.7
77GLM-4.783.3% ±2.4
77Qwen3.6 Flash83.3% ±2.7
80GPT-5.5 Instant82.5% ±2.3
81Qwen 3.5 Flash (hosted 35B-A3B)82.3% ±2.7
82Qwen3.7 Plusnone effort81.8% ±2.7
82Gemini 3.1 Flash-Litehigh effort81.8% ±2.7
82o3high effort81.8% ±2.1
85MiniMax M3none effort81.3% ±2.8
86Qwen3.5-35B-A3Bnone effort81.2% ±2.3
87Qwen3.7 Flashnone effort80.8% ±2.8
88Claude Sonnet 5max effort80.3% ±2.8
88Claude Sonnet 4.580.3% ±2.9
90Qwen3-235B-A22B (Jul 2025)80.1% ±2.6
91o4-minihigh effort79.6% ±2.4
92Qwen3.5 9Bnone effort78.9% ±2.4
93Claude Sonnet 4.6max effort78.8% ±2.9
94GPT-5.4 nanohigh effort78.5% ±2.4
95Doubao Seed 1.677.6% ±2.5
96DeepSeek V3.277.3% ±2.0
97Claude Sonnet 477.0% ±2.9
97o3-minihigh effort77.0% ±2.6
99Claude Opus 4.176.8% ±3.0
99o1high effort76.8% ±3.0
99Claude 3.7 Sonnet76.8% ±3.0
102DeepSeek-R1 (May 2025)76.3% ±2.4
103Gemma 4 31Bminimal effort75.8% ±3.1
103GPT-OSS 120Bhigh effort75.8% ±2.7
103Grok 3 [Beta]75.8% ±2.7
106Grok 3 Minihigh effort75.5% ±2.8
107GPT-5 minihigh effort75.0% ±2.3
108Gemma 4 26B A4Bminimal effort73.2% ±3.2
109Claude Opus 472.7% ±3.0
110Qwen3 Max72.6% ±2.7
111DeepSeek-R171.7% ±3.1
112Seed OSS 36B Instruct71.5% ±2.7
113Qwen3 235B-A22B70.7% ±2.7
114GPT-5 nanohigh effort69.4% ±2.8
115GPT-4.568.7% ±3.3
116DeepSeek-V3 (Mar 2025)67.6% ±2.7
117Llama 4 Maverick67.0% ±2.8
118GPT-4.166.9% ±2.8
119Gemini 2.5 Pro (May 2025)66.7% ±3.4
120Claude Haiku 4.565.8% ±3.2
120GPT-4.1 mini65.8% ±2.7
122Gemini 2.0 Pro65.7% ±3.4
123QWQ Plus65.4% ±2.8
124QwQ-32B65.3% ±2.8
125Gemini 2.0 Flash64.1% ±2.7
125DeepSeek R1 Distill Qwen 32B64.1% ±2.8
127Qwen3-30B-A3B (Jul 2025)62.8% ±2.8
128o1-minihigh effort62.4% ±2.7
129Mistral Medium 359.5% ±2.8
130Gemini 2.0 Flash Thinking57.1% ±3.5
131DeepSeek V356.5% ±2.8
132Qwen2.5-Max56.1% ±2.8
132Phi-456.1% ±2.6
132Magistral Small 1.056.1% ±3.5
135R1 Distill Llama 70B55.7% ±3.0
136Claude 3.5 Sonnet (October 2024)55.3% ±2.8
137Qwen3 32Bnone effort54.1% ±2.8
138Claude 3.5 Sonnet54.0% ±2.8
139Grok 253.8% ±2.7
140Qwen3 14Bnone effort53.4% ±2.8
141Llama 4 Scout51.8% ±3.2
142Gemini 1.5 Pro51.5% ±2.6
143Llama 3.1 405B50.9% ±2.6
144Qwen3 30B A3Bnone effort50.4% ±2.8
145o1-preview50.3% ±2.8
146Mistral Large 250.2% ±2.7
147Qwen2.5-72B49.1% ±2.7
147Mistral Small 3.249.1% ±2.8
149GPT-4.1 nano48.9% ±2.5
149GPT-4o48.9% ±2.6
151Qwen Plus48.1% ±2.7
152Gemma 3 27B47.7% ±2.7
153Magistral Small 1.247.6% ±2.6
154Qwen3 8Bnone effort47.5% ±2.6
155Llama 3.3 70B47.4% ±2.9
156Claude 3 Opus47.2% ±2.6
157GPT-4 Turbo (Apr 2024)46.6% ±2.7
158Mistral Small 346.3% ±2.5
158Tulu 3 (Tülu 3) 70B46.3% ±2.4
160Qwen2.5 32B46.1% ±2.6
161GPT-OSS 20Bhigh effort46.0% ±3.6
162Qwen3 4B Instruct45.8% ±2.9
163GLM-4.7-Flashnone effort45.1% ±2.2
164Mistral Small 3.144.7% ±2.3
164DeepSeek R1 Distill Qwen 14B44.7% ±2.9
166Llama 3.1 70B44.2% ±2.5
167Gemini 1.5 Flash43.8% ±2.4
168Qwen3 4Bnone effort43.6% ±2.7
169WizardLM-2 8x22B43.4% ±2.3
170GPT-4 Turbo (Nov 2023)42.3% ±2.4
171Qwen Turbo41.8% ±2.6
172Llama 3.2 90B41.0% ±2.3
173Qwen2 72B40.8% ±2.6
174Claude 3 Sonnet40.6% ±2.4
174Llama 3 70B40.6% ±2.6
176Gemma 3 12B39.5% ±2.5
177Mistral Large38.8% ±2.5
178Claude 3.5 Haiku38.1% ±2.6
179GPT-4o mini37.7% ±2.4
180Hermes 2 Theta Llama-3 70B37.5% ±2.5
181Gemma 2 27B36.5% ±2.4
182Claude 3 Haiku36.3% ±2.4
183GPT-4 (Mar 2023)35.7% ±2.4
184Qwen2.5 7B35.5% ±2.3
185Claude 234.7% ±2.5
186Mixtral 8x22B34.1% ±2.1
187Gemini 1.0 Pro34.0% ±2.0
188Eurus 2 7B PRIME33.9% ±2.1
189DeepSeek R1 Distill Qwen 1.5B33.6% ±2.1
190Claude 2.133.0% ±2.3
190Gemini 1.5 Flash 8B33.0% ±2.4
192DBRX32.9% ±3.3
193Yi 1.5 34B32.0% ±2.0
194Qwen1.5 32B30.7% ±2.0
194GPT-4 (Jun 2023)30.7% ±2.3
196Qwen3 1.7Bnone effort30.6% ±2.3
197Mixtral 8x7B30.2% ±1.9
198Mistral Nemo29.9% ±2.1
199Qwen1.5 72B28.8% ±2.0
200Granite 4.0 Micro28.3% ±1.8
201GPT-3.5 Turbo27.6% ±2.0
201phi-3-medium 14B27.6% ±1.6
203Gemma 2 9B27.5% ±2.3
204Ministral 8B27.1% ±2.1
205Llama 3.1 8B27.0% ±1.9
206Llama 2-70B26.3% ±2.0
207Llama 3-8B26.1% ±1.7
208Ministral 3B25.3% ±1.9
209DeepSeek LLM 67B24.6% ±1.4
210Granite-4.0-1B24.0% ±1.6
211Llama 3.2 1B23.9% ±1.6
212Gemma 3 4B23.2% ±1.9
213Gemma 3 1B19.9% ±1.5
214Yi 34B14.7% ±1.1
215Mistral 7B14.2% ±1.0
216Granite-4.0-350M11.2% ±0.9
217DeepSeek R1 Qwen3 8B9.3% ±1.1