Knowledge & reasoning

GPQA Diamond leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on GPQA Diamond is 96.0% by GPT-6 Astra. 177 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 4 Argon	97.1%	estimated ± 2.2 pp, medium confidence
2	Claude Sonnet 5.5	96.0%	estimated ± 1.0 pp, low confidence
3	Claude Opus 5.5	96.0%	estimated ± 1.0 pp, low confidence
4	Ling 3.1 Flash	96.0%	estimated ± 1.0 pp, low confidence
5	GPT-6 Astra	96.0%	measured
6	GPT-6.1 Sol	96.0%	estimated ± 1.0 pp, medium confidence
7	Sakana Fugu	95.5%	measured
8	Sakana Fugu-Ultra	95.5%	measured
9	Claude Fable 5	95.2%	estimated ± 1.4 pp, high confidence
10	Gemini 3.8 Flash	95.2%	estimated ± 1.4 pp, high confidence
11	Pareto 26.9	95.0%	estimated ± 1.9 pp, high confidence
12	GPT-6 Luna	94.9%	estimated ± 1.0 pp, medium confidence
13	GPT-6 Sol	94.9%	estimated ± 1.0 pp, medium confidence
14	Claude Fable 5.1	94.7%	estimated ± 1.4 pp, high confidence
15	MiMo-V2.6-Pro	94.7%	estimated ± 2.2 pp, high confidence
16	GPT-5.6 Sol	94.6%	measured
17	Muse Spark 1.3	94.5%	estimated ± 2.2 pp, high confidence
18	Gemini 3.1 Pro	94.3%	measured
19	Claude Opus 4.7 (Adaptive)	94.2%	measured
20	Claude Mythos 5	94.1%	estimated ± 0.4 pp, high confidence
21	Claude Opus 5	94.0%	estimated ± 1.0 pp, medium confidence
22	Claude Haiku 5.5	93.9%	estimated ± 1.9 pp, high confidence
23	Gemini 3.7 Flash	93.8%	estimated ± 1.4 pp, high confidence
24	dots3-note Preview	93.7%	estimated ± 1.5 pp, high confidence
25	Claude Opus 4.8	93.6%	measured
26	GPT-5.5	93.6%	measured
27	Muse Spark 1.1	93.6%	estimated ± 1.0 pp, medium confidence
28	GPT-5.5 Pro	93.5%	estimated ± 1.4 pp, high confidence
29	Kimi K3	93.5%	measured
30	Step 5 Preview	93.5%	measured
31	GPT-5.4 Pro	93.3%	estimated ± 1.4 pp, high confidence
32	GPT-5.6 Terra	92.9%	measured
33	GPT-5.4	92.8%	measured
34	Ornith-1.5-397B	92.8%	measured
35	Gemini 3.5 Flash	92.7%	measured
36	Grok 4.7	92.7%	estimated ± 1.0 pp, medium confidence
37	Qwen3.8 Max	92.6%	measured
38	Claude Opus 4.6 (Adaptive)	92.6%	estimated ± 1.4 pp, high confidence
39	Qwen3.8 Max Preview	92.5%	estimated ± 2.2 pp, high confidence
40	Pareto 26.10 Preview	92.4%	measured
41	Qwen3.7 Max	92.4%	measured
42	GPT-5.2	92.4%	estimated ± 0.4 pp, high confidence
43	GPT-5.6 Luna	92.3%	measured
44	Hy4 preview	92.3%	measured
45	GPT-5.3 Codex	92.3%	estimated ± 2.2 pp, high confidence
46	MiniMax M3	92.1%	estimated ± 1.7 pp, high confidence
47	Muse Spark 1.2	91.8%	estimated ± 2.0 pp, high confidence
48	Gemini 3.6 Flash	91.8%	estimated ± 1.4 pp, high confidence
49	Qwen3.8-Flash-Next	91.7%	measured
50	Agents-A1	91.4%	estimated ± 2.3 pp, high confidence
51	GLM-5.2	91.2%	measured
52	Qwen3.8-Omni-Flash	91.0%	measured
53	DeepSeek V4.1 Flash	90.9%	measured
54	Claude Opus 4.7	90.8%	estimated ± 1.7 pp, high confidence
55	Kimi K2.6	90.5%	measured
56	Beam	90.5%	measured
57	Qwen3.6 Plus	90.4%	estimated ± 0.1 pp, high confidence
58	Qwen3.7 Plus	90.3%	measured
59	Grok 4.3	90.2%	estimated ± 0.1 pp, high confidence
60	DeepSeek V4 Pro 0813	90.1%	measured
61	Claude Sonnet 5	90.1%	estimated ± 1.7 pp, high confidence
62	Grok 4.6	90.0%	estimated ± 1.4 pp, high confidence
63	Interfaze Beta	89.9%	measured
64	Claude Sonnet 4.6	89.9%	estimated ± 0.1 pp, high confidence
65	GLM-5.3	89.6%	estimated ± 1.7 pp, high confidence
66	Grok 4.5	89.5%	estimated ± 1.4 pp, high confidence
67	GPT-5.2-Codex	89.5%	estimated ± 2.2 pp, high confidence
68	Gemini 3 Flash	89.5%	estimated ± 1.7 pp, high confidence
69	Inkling-Small	89.5%	measured
70	Muse Spark	89.5%	measured
71	Gemini 3 Pro Deep Think	89.4%	estimated ± 1.5 pp, high confidence
72	MiMo-V2.6-Flash	89.3%	estimated ± 2.2 pp, high confidence
73	Kimi K2.7 Code	89.2%	estimated ± 2.2 pp, high confidence
74	Mistral Large 4	89.2%	estimated ± 2.2 pp, high confidence
75	Claude Opus 4.6	89.2%	measured
76	Ornith-1.5-35B-A3B	89.2%	measured
77	Qwen3.8-27B	89.2%	measured
78	Solar Pro 4	89.0%	measured
79	Apodex 1.1	88.8%	estimated ± 2.2 pp, high confidence
80	Apodex 1.1 Mini	88.8%	estimated ± 2.2 pp, high confidence
81	GLM-5.3-Flash	88.7%	estimated ± 1.7 pp, high confidence
82	Hy3	88.5%	estimated ± 2.2 pp, high confidence
83	Grok 4.20	88.5%	measured
84	Qwen3.5 397B	88.4%	estimated ± 0.4 pp, high confidence
85	DeepSeek V4 Flash 0731	88.1%	measured
86	GPT-5.4 mini	88.0%	estimated ± 0.1 pp, high confidence
87	Inkling	87.9%	measured
88	Qwen3.6-27B	87.8%	estimated ± 0.4 pp, high confidence
89	Kimi K2.5	87.6%	measured
90	Claude Opus 4.5 Thinking	87.6%	estimated ± 1.4 pp, medium confidence
91	Kimi K2.5 (Reasoning)	87.6%	estimated ± 0.4 pp, high confidence
92	Qwen 3.6 Max (preview)	87.3%	estimated ± 2.2 pp, high confidence
93	Hy3 Preview	87.2%	measured
94	Gemini 3.5 Flash-Lite	87.2%	estimated ± 1.7 pp, high confidence
95	MiMo-V2-Pro	87.1%	estimated ± 2.2 pp, high confidence
96	MiniMax M2.7	87.0%	measured
97	Nemotron 3 Ultra	87.0%	measured
98	Claude Opus 4.5	87.0%	estimated ± 0.4 pp, high confidence
99	Qwen3.5-122B-A10B	86.6%	estimated ± 0.4 pp, high confidence
100	Qwen3.5 Flash	86.6%	estimated ± 1.7 pp, medium confidence
101	MiMo-V2.5-Pro	86.4%	estimated ± 1.7 pp, medium confidence
102	Ornith-1.5-9B	86.4%	measured
103	Solar Open 2	86.3%	measured
104	Gemini 3 Pro	86.3%	estimated ± 1.4 pp, medium confidence
105	GLM-5.1	86.2%	measured
106	Seed 2.1 Pro	86.2%	estimated ± 12.9 pp, low confidence
107	GPT-5 (high)	86.1%	estimated ± 2.2 pp, high confidence
108	o3-pro	86.0%	estimated ± 2.4 pp, high confidence
109	GLM-5	86.0%	measured
110	Qwen3.6-35B-A3B	86.0%	estimated ± 0.4 pp, high confidence
111	MiMo-V2.5	85.8%	estimated ± 1.7 pp, medium confidence
112	GPT-5.1	85.8%	estimated ± 1.4 pp, medium confidence
113	Ternary Bonsai 2 27B	85.8%	measured
114	GLM-5-Turbo	85.8%	estimated ± 2.2 pp, high confidence
115	GLM-4.7	85.6%	estimated ± 0.1 pp, high confidence
116	A.X K2	85.6%	measured
117	Gemini 3.1 Flash-Lite	85.5%	estimated ± 1.7 pp, medium confidence
118	Qwen3.5-27B	85.5%	estimated ± 0.4 pp, high confidence
119	Grok 4	85.2%	estimated ± 2.2 pp, high confidence
120	Ling 3.0 Flash	85.0%	measured
121	GPT-5.1-Codex	84.6%	estimated ± 2.2 pp, high confidence
122	GPT-5.1-Codex-Max	84.6%	estimated ± 2.2 pp, high confidence
123	GPT-5 (medium)	84.5%	estimated ± 2.2 pp, high confidence
124	Claude Sonnet 4.5 Thinking	84.3%	estimated ± 1.4 pp, medium confidence
125	Gemma 4 31B	84.3%	estimated ± 0.4 pp, high confidence
126	MAI-Thinking-1	84.2%	measured
127	Qwen3.5-35B-A3B	84.2%	estimated ± 0.4 pp, high confidence
128	Ling 3.0 Flash FP8	84.0%	measured
129	Seed 2.1 Turbo	83.9%	estimated ± 12.9 pp, low confidence
130	Claude Sonnet 4.5	83.4%	estimated ± 0.4 pp, high confidence
131	MiMo-V2-Flash	83.3%	estimated ± 0.1 pp, high confidence
132	Claude 4.1 Opus	83.0%	estimated ± 2.4 pp, high confidence
133	Gemini 2.5 Pro	83.0%	estimated ± 0.4 pp, high confidence
134	GPT-5.4 nano	82.8%	estimated ± 0.1 pp, high confidence
135	MiMo-V2-Omni	82.6%	estimated ± 2.2 pp, high confidence
136	Ling 3.0 Flash VL	82.5%	estimated ± 2.2 pp, high confidence
137	Muse Glimmer 30B	82.5%	estimated ± 2.2 pp, high confidence
138	K-EXAONE 2.0	82.2%	measured
139	Step 3.7 Flash	82.2%	estimated ± 2.2 pp, high confidence
140	Nemotron 3 Super 100B	81.8%	estimated ± 2.2 pp, high confidence
141	GLM-4.6	81.4%	estimated ± 1.7 pp, medium confidence
142	o3	81.4%	estimated ± 2.2 pp, high confidence
143	Qwen3.5 397B (Reasoning)	81.2%	estimated ± 2.2 pp, high confidence
144	GPT-OSS 120B	81.0%	estimated ± 2.2 pp, high confidence
145	Grok 4.1 Fast (Reasoning)	80.9%	estimated ± 2.2 pp, high confidence
146	Grok 4 Fast (Reasoning)	80.7%	estimated ± 2.2 pp, high confidence
147	Quasar 438B	80.5%	estimated ± 2.2 pp, high confidence
148	Claude Haiku 4.5	79.9%	estimated ± 1.7 pp, medium confidence
149	GLM-4.5	79.9%	estimated ± 1.7 pp, medium confidence
150	Gemma 4 26B A4B	79.7%	estimated ± 1.9 pp, high confidence
151	GLM-5V-Turbo	79.4%	estimated ± 2.2 pp, high confidence
152	Mercury 2.5	79.0%	measured
153	o1-pro	79.0%	estimated ± 0.4 pp, high confidence
154	Gemma 4 12B	78.8%	measured
155	DeepSeek-R1	78.5%	estimated ± 2.2 pp, high confidence
156	Qwen3 235B 2507	77.5%	estimated ± 0.4 pp, high confidence
157	DeepSeek V3.1 (Reasoning)	77.4%	estimated ± 2.2 pp, high confidence
158	o3-mini	77.2%	estimated ± 0.4 pp, high confidence
159	K-Exaone	77.1%	estimated ± 2.2 pp, high confidence
160	Trinity-Large-Thinking	76.3%	measured
161	Claude 4.1 Opus Thinking	76.1%	estimated ± 2.2 pp, high confidence
162	Command A+	75.7%	estimated ± 2.2 pp, high confidence
163	o1	75.7%	estimated ± 0.4 pp, high confidence
164	Qwen3 Max	75.6%	estimated ± 2.2 pp, high confidence
165	Nemotron 3.5 Lightning 30B A3B NVFP4	75.6%	measured
166	Nemotron 3 Nano 30B	75.2%	estimated ± 2.2 pp, high confidence
167	DeepSeek V3.2	75.1%	estimated ± 2.2 pp, high confidence
168	North Mini Code	75.0%	estimated ± 2.2 pp, high confidence
169	GPT-OSS 20B	74.9%	estimated ± 2.2 pp, high confidence
170	Sarvam 105B	74.9%	estimated ± 2.2 pp, high confidence
171	Solar Pro 3	74.3%	estimated ± 2.2 pp, high confidence
172	Mistral Small 4	74.0%	estimated ± 2.2 pp, high confidence
173	Mistral Small 4 (Reasoning)	74.0%	estimated ± 2.2 pp, high confidence
174	Ling 3.0 Tiny	73.5%	estimated ± 2.2 pp, high confidence
175	o1-preview	72.6%	estimated ± 2.4 pp, high confidence
176	Grok Code Fast 1	72.4%	estimated ± 2.2 pp, high confidence
177	Nemotron 3 Nano Omni 30B A3B	72.2%	measured
178	Qwen3-Omni-30B-A3B-Thinking	71.9%	estimated ± 2.2 pp, high confidence
179	Sarvam 30B	71.9%	estimated ± 2.2 pp, high confidence
180	Kimi K2	71.9%	estimated ± 2.2 pp, high confidence
181	Nemotron Ultra 253B	71.9%	estimated ± 2.2 pp, high confidence
182	GLM-4.5-Air	71.5%	estimated ± 2.2 pp, high confidence
183	LFM2.5-8B-A1B	71.4%	estimated ± 2.2 pp, high confidence
184	Celeris-1	71.3%	estimated ± 2.2 pp, high confidence
185	DeepSeek V3.1	71.2%	estimated ± 2.2 pp, high confidence
186	ZAYA1-8B	71.0%	measured
187	Granite-4.0-H-350M	71.0%	estimated ± 2.2 pp, high confidence
188	LFM2.5-2.6B	70.8%	estimated ± 2.2 pp, high confidence
189	Exaone 4.0 1.2B	70.3%	estimated ± 2.2 pp, high confidence
190	MiniCPM5-2B	70.2%	measured
191	Granite-4.0-350M	70.1%	estimated ± 2.2 pp, high confidence
192	Grok 4.1 Fast	69.7%	estimated ± 2.2 pp, high confidence
193	LFM2.5-VL-1.6B-Extract	69.7%	estimated ± 2.2 pp, high confidence
194	Exaone 4.0 32B	69.6%	estimated ± 2.2 pp, high confidence
195	Granite-4.0-H-1B	69.6%	estimated ± 2.2 pp, high confidence
196	Phi-4 Multimodal Instruct	69.6%	estimated ± 2.2 pp, high confidence
197	Llama 4 Maverick	69.5%	estimated ± 2.2 pp, high confidence
198	LongCat-Flash-Lite-Sparse	69.5%	measured
199	DeepSeek V3 0324	69.3%	estimated ± 2.2 pp, medium confidence
200	Gemini 2.5 Flash	69.3%	estimated ± 2.2 pp, medium confidence
201	DeepSeek R1 Distill Qwen 32B	69.2%	estimated ± 2.2 pp, medium confidence
202	Gemini 1.5 Pro	69.2%	estimated ± 2.2 pp, medium confidence
203	Qwen3-Omni-30B-A3B-Instruct	69.2%	estimated ± 2.2 pp, medium confidence
204	Gemma 3 27B	69.0%	estimated ± 2.2 pp, medium confidence
205	Claude 4 Sonnet	69.0%	estimated ± 2.2 pp, medium confidence
206	Gemini 1.0 Pro	68.9%	estimated ± 2.2 pp, medium confidence
207	GPT-4o mini	68.9%	estimated ± 2.2 pp, medium confidence
208	Mistral Large 3	68.9%	estimated ± 2.2 pp, medium confidence
209	Claude 3 Haiku	68.8%	estimated ± 2.2 pp, medium confidence
210	Mistral Medium 3	68.8%	estimated ± 2.2 pp, medium confidence
211	Llama 3.1 405B	68.7%	estimated ± 2.2 pp, medium confidence
212	Llama 4 Scout	68.5%	estimated ± 2.2 pp, medium confidence
213	Phi-4	68.5%	estimated ± 2.2 pp, medium confidence
214	Solar Pro 2	68.4%	estimated ± 2.2 pp, medium confidence
215	Ultravox v0.6 Llama 3.3 70B	68.3%	estimated ± 2.2 pp, medium confidence
216	Qwen2.5 Coder 32B Instruct	68.2%	estimated ± 2.2 pp, medium confidence
217	Mistral Large 2	68.0%	estimated ± 2.2 pp, medium confidence
218	Nova Pro	67.9%	estimated ± 2.2 pp, medium confidence
219	GPT-4 Turbo	67.7%	estimated ± 2.2 pp, medium confidence
220	Claude 3 Opus	67.4%	estimated ± 2.2 pp, medium confidence
221	Laguna XS.2	67.4%	estimated ± 1.7 pp, medium confidence
222	GPT-4o	67.0%	estimated ± 2.2 pp, medium confidence
223	Granite 4.2 30B	66.4%	estimated ± 0.4 pp, high confidence
224	GPT-4.1	66.3%	estimated ± 0.4 pp, high confidence
225	GPT-4.1 mini	64.2%	estimated ± 0.4 pp, high confidence
226	Granite 4.2 8B	64.1%	estimated ± 0.4 pp, high confidence
227	Trinity-Large-Preview	63.3%	measured
228	Claude 3.5 Sonnet	59.4%	estimated ± 0.4 pp, high confidence
229	DeepSeek V3	59.1%	estimated ± 0.4 pp, high confidence
230	Ling 2.6 Flash	59.0%	estimated ± 0.4 pp, high confidence
231	Gemma 4 E4B	58.6%	estimated ± 0.4 pp, high confidence
232	Mellum2-12B-A2.5B-Thinking	57.6%	measured
233	ZAYA1-74B-Preview	57.3%	measured
234	Granite 4.2 3B	54.8%	estimated ± 0.4 pp, high confidence
235	GPT-4.1 nano	50.3%	estimated ± 0.4 pp, high confidence
236	Mistral Medium 3.5 128B	48.6%	estimated ± 1.7 pp, medium confidence
237	LLaDA2.2-mini	44.4%	measured
238	Gemma 4 E2B	43.4%	estimated ± 0.4 pp, high confidence
239	Soofi S 30B-A3B	43.4%	measured
240	Mellum2-12B-A2.5B-Instruct	40.9%	measured
241	Laguna M.1	39.9%	estimated ± 1.7 pp, medium confidence
242	MiniCPM5-1B	26.3%	measured
243	LFM2.5-VL-450M	25.7%	estimated ± 0.4 pp, high confidence
244	LFM2.5-230M	25.4%	measured

GPQA Diamond leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General