Knowledge & reasoning

GPQA leaderboard

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

As of 2026-10-07, the highest measured score on GPQA is 96.0% by GPT-6 Astra. 160 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Sonnet 5.5	96.0%	estimated ± 1.0 pp, low confidence
2	Claude Opus 5.5	96.0%	estimated ± 1.0 pp, low confidence
3	Ling 3.1 Flash	96.0%	estimated ± 1.0 pp, low confidence
4	GPT-6 Astra	96.0%	measured
5	GPT-6.1 Sol	95.9%	estimated ± 0.8 pp, high confidence
6	Claude Fable 5.1	95.9%	estimated ± 1.6 pp, medium confidence
7	Claude Opus 5	95.6%	estimated ± 0.8 pp, high confidence
8	Sakana Fugu	95.5%	measured
9	Sakana Fugu-Ultra	95.5%	measured
10	Claude Fable 5	95.1%	estimated ± 1.6 pp, medium confidence
11	Gemini 3.8 Flash	94.8%	estimated ± 0.8 pp, high confidence
12	Pareto 26.9	94.8%	estimated ± 2.1 pp, high confidence
13	GPT-5.6 Sol	94.6%	measured
14	Gemini 3.1 Pro	94.2%	estimated ± 0.1 pp, high confidence
15	Claude Opus 4.7 (Adaptive)	94.2%	measured
16	GPT-6 Sol	94.2%	estimated ± 0.8 pp, high confidence
17	Claude Mythos 5	94.1%	measured
18	GPT-5.5 Pro	93.8%	estimated ± 1.6 pp, high confidence
19	Gemini 3.7 Flash	93.7%	estimated ± 1.6 pp, medium confidence
20	GPT-5.4 Pro	93.7%	estimated ± 1.6 pp, high confidence
21	Grok 4.6	93.7%	estimated ± 0.8 pp, high confidence
22	Claude Haiku 5.5	93.6%	estimated ± 2.1 pp, high confidence
23	Claude Opus 4.8	93.6%	measured
24	GPT-5.5	93.6%	measured
25	Muse Spark 1.1	93.6%	estimated ± 1.0 pp, medium confidence
26	dots3-note Preview	93.6%	estimated ± 1.6 pp, high confidence
27	Claude Opus 4.7	93.5%	estimated ± 1.6 pp, medium confidence
28	Step 5 Preview	93.5%	estimated ± 0.4 pp, high confidence
29	Kimi K3	93.5%	measured
30	Grok 4.5	93.2%	estimated ± 0.8 pp, high confidence
31	GPT-6 Luna	93.1%	estimated ± 0.8 pp, medium confidence
32	Gemini 3.6 Flash	93.0%	estimated ± 1.6 pp, high confidence
33	GPT-5.6 Terra	92.9%	measured
34	Gemini 4 Argon	92.8%	estimated ± 3.2 pp, high confidence
35	GPT-5.4	92.8%	measured
36	Ornith-1.5-397B	92.8%	measured
37	Grok 4.7	92.7%	estimated ± 1.0 pp, medium confidence
38	Qwen3.8 Max	92.6%	measured
39	Muse Spark 1.3	92.6%	estimated ± 3.2 pp, high confidence
40	MiMo-V2.6-Pro	92.5%	estimated ± 3.2 pp, high confidence
41	Claude Opus 4.6 (Adaptive)	92.4%	estimated ± 1.6 pp, high confidence
42	Pareto 26.10 Preview	92.4%	estimated ± 0.4 pp, high confidence
43	Qwen3.8 Max Preview	92.4%	estimated ± 3.2 pp, high confidence
44	GPT-5.2	92.4%	measured
45	Qwen3.7 Max	92.4%	measured
46	Gemini 3 Flash	92.3%	estimated ± 1.6 pp, high confidence
47	GPT-5.6 Luna	92.3%	measured
48	Hy4 preview	92.3%	measured
49	Gemini 3.5 Flash	92.2%	measured
50	Apodex 1.1	92.1%	estimated ± 2.2 pp, high confidence
51	Muse Spark 1.2	92.0%	estimated ± 1.6 pp, high confidence
52	Mistral Large 4	91.8%	estimated ± 3.2 pp, high confidence
53	Qwen3.8-Flash-Next	91.7%	measured
54	MiMo-V2.6-Flash	91.7%	estimated ± 3.2 pp, high confidence
55	Claude Opus 4.6	91.3%	measured
56	Agents-A1	91.3%	estimated ± 2.2 pp, high confidence
57	Claude Sonnet 5	91.3%	estimated ± 1.6 pp, high confidence
58	GLM-5.2	91.2%	measured
59	Qwen3.8-Omni-Flash	91.0%	measured
60	DeepSeek V4.1 Flash	90.9%	measured
61	GPT-5.3 Codex	90.8%	estimated ± 3.2 pp, high confidence
62	GLM-5.3	90.6%	estimated ± 1.6 pp, high confidence
63	Beam	90.5%	estimated ± 0.4 pp, high confidence
64	Kimi K2.6	90.5%	measured
65	Qwen3.6 Plus	90.4%	measured
66	Qwen3.7 Plus	90.3%	measured
67	DeepSeek V4 Pro 0813	90.1%	measured
68	Grok 4.3	90.1%	measured
69	Gemini 3.1 Flash-Lite	90.0%	estimated ± 1.6 pp, high confidence
70	GLM-5.3-Flash	89.9%	estimated ± 1.6 pp, high confidence
71	Claude Sonnet 4.6	89.9%	measured
72	Interfaze Beta	89.9%	measured
73	Gemini 3 Pro Deep Think	89.9%	estimated ± 1.6 pp, high confidence
74	MiMo-V2-Pro	89.9%	estimated ± 3.2 pp, high confidence
75	GPT-5.2-Codex	89.8%	estimated ± 3.2 pp, high confidence
76	Gemini 3.5 Flash-Lite	89.6%	estimated ± 1.6 pp, high confidence
77	Muse Spark	89.5%	estimated ± 0.1 pp, high confidence
78	Inkling-Small	89.5%	measured
79	Qwen 3.6 Max (preview)	89.4%	estimated ± 2.6 pp, high confidence
80	Quasar 438B	89.2%	estimated ± 3.2 pp, high confidence
81	Ornith-1.5-35B-A3B	89.2%	measured
82	Qwen3.8-27B	89.2%	measured
83	GLM-5-Turbo	89.2%	estimated ± 3.2 pp, high confidence
84	Apodex 1.1 Mini	89.1%	estimated ± 3.2 pp, high confidence
85	Solar Pro 4	89.0%	estimated ± 0.4 pp, high confidence
86	Claude Opus 4.5 Thinking	88.9%	estimated ± 1.6 pp, high confidence
87	Kimi K2.7 Code	88.8%	estimated ± 3.2 pp, high confidence
88	Hy3	88.6%	estimated ± 3.2 pp, high confidence
89	Grok 4.20	88.5%	estimated ± 0.1 pp, high confidence
90	MiMo-V2.5-Pro	88.5%	estimated ± 1.6 pp, high confidence
91	Qwen3.5 397B	88.4%	measured
92	Ling 3.0 Flash VL	88.3%	estimated ± 3.2 pp, high confidence
93	Seed 2.1 Pro	88.3%	estimated ± 2.6 pp, high confidence
94	MiniMax M3	88.1%	estimated ± 1.6 pp, high confidence
95	DeepSeek V4 Flash 0731	88.1%	measured
96	GPT-5.1	88.1%	estimated ± 1.7 pp, high confidence
97	Qwen3.5 Flash	88.0%	estimated ± 1.6 pp, high confidence
98	GPT-5.4 mini	88.0%	measured
99	Gemini 3 Pro	88.0%	estimated ± 1.6 pp, high confidence
100	MiMo-V2-Omni	88.0%	estimated ± 3.2 pp, high confidence
101	Inkling	87.9%	measured
102	GPT-5.1-Codex	87.8%	estimated ± 3.2 pp, high confidence
103	GPT-5.1-Codex-Max	87.8%	estimated ± 3.2 pp, high confidence
104	Qwen3.6-27B	87.8%	measured
105	GLM-5V-Turbo	87.7%	estimated ± 3.2 pp, high confidence
106	Kimi K2.5	87.6%	measured
107	Kimi K2.5 (Reasoning)	87.6%	measured
108	GPT-5 (high)	87.4%	estimated ± 3.2 pp, high confidence
109	GPT-5 (medium)	87.4%	estimated ± 3.2 pp, high confidence
110	Claude 4.1 Opus Thinking	87.3%	estimated ± 3.2 pp, high confidence
111	Hy3 Preview	87.2%	measured
112	Command A+	87.1%	estimated ± 3.2 pp, high confidence
113	Grok 4	87.1%	estimated ± 3.2 pp, high confidence
114	MiniMax M2.7	87.0%	estimated ± 0.1 pp, high confidence
115	Claude Opus 4.5	87.0%	measured
116	Nemotron 3 Ultra	87.0%	measured
117	MiMo-V2.5	86.9%	estimated ± 1.6 pp, high confidence
118	o3-pro	86.7%	estimated ± 3.2 pp, high confidence
119	Qwen3.5-122B-A10B	86.6%	measured
120	Seed 2.1 Turbo	86.6%	estimated ± 2.6 pp, high confidence
121	Qwen3.5 397B (Reasoning)	86.4%	estimated ± 3.2 pp, high confidence
122	Ornith-1.5-9B	86.4%	measured
123	Solar Open 2	86.3%	estimated ± 0.4 pp, high confidence
124	GLM-5.1	86.2%	estimated ± 0.1 pp, high confidence
125	GLM-4.6	86.2%	estimated ± 1.6 pp, high confidence
126	GLM-5	86.0%	measured
127	Qwen3.6-35B-A3B	86.0%	measured
128	Ternary Bonsai 2 27B	85.8%	measured
129	GLM-4.7	85.7%	measured
130	A.X K2	85.6%	estimated ± 0.4 pp, high confidence
131	Grok 4.1 Fast (Reasoning)	85.6%	estimated ± 3.2 pp, high confidence
132	Qwen3.5-27B	85.5%	measured
133	o3	85.4%	estimated ± 3.2 pp, high confidence
134	GLM-4.5	85.2%	estimated ± 1.6 pp, high confidence
135	Ling 3.0 Flash	85.0%	measured
136	Claude Sonnet 4.5 Thinking	84.8%	estimated ± 1.6 pp, high confidence
137	Step 3.7 Flash	84.8%	estimated ± 3.2 pp, high confidence
138	Gemma 4 31B	84.3%	measured
139	MAI-Thinking-1	84.2%	measured
140	Qwen3.5-35B-A3B	84.2%	measured
141	Ling 3.0 Flash FP8	84.0%	measured
142	Claude 4.1 Opus	83.8%	estimated ± 3.2 pp, high confidence
143	MiMo-V2-Flash	83.7%	measured
144	Claude Sonnet 4.5	83.4%	measured
145	Grok 4 Fast (Reasoning)	83.1%	estimated ± 3.2 pp, high confidence
146	Gemini 2.5 Pro	83.0%	measured
147	Claude Haiku 4.5	82.9%	estimated ± 1.6 pp, high confidence
148	GPT-5.4 nano	82.8%	measured
149	Muse Glimmer 30B	82.5%	estimated ± 3.2 pp, high confidence
150	K-EXAONE 2.0	82.2%	estimated ± 0.4 pp, high confidence
151	Claude 4 Sonnet	81.3%	estimated ± 3.2 pp, high confidence
152	DeepSeek V3.2	80.4%	estimated ± 3.2 pp, high confidence
153	Gemma 4 26B A4B	80.2%	estimated ± 2.1 pp, high confidence
154	Qwen3 Max	79.7%	estimated ± 3.2 pp, high confidence
155	Mistral Medium 3.5 128B	79.6%	estimated ± 1.6 pp, medium confidence
156	Mercury 2.5	79.0%	estimated ± 0.4 pp, high confidence
157	o1-pro	79.0%	measured
158	Gemma 4 12B	78.8%	measured
159	Qwen3 235B 2507	77.5%	measured
160	K-Exaone	77.2%	estimated ± 3.2 pp, high confidence
161	o3-mini	77.2%	measured
162	Grok Code Fast 1	76.5%	estimated ± 3.2 pp, high confidence
163	Trinity-Large-Thinking	76.3%	estimated ± 0.4 pp, high confidence
164	o1	75.7%	measured
165	DeepSeek V3.1	75.7%	estimated ± 3.2 pp, high confidence
166	Nemotron 3.5 Lightning 30B A3B NVFP4	75.6%	measured
167	DeepSeek V3.1 (Reasoning)	75.1%	estimated ± 3.2 pp, high confidence
168	DeepSeek-R1	74.2%	estimated ± 3.2 pp, high confidence
169	Laguna XS.2	73.7%	estimated ± 1.6 pp, medium confidence
170	Laguna M.1	73.4%	estimated ± 1.6 pp, medium confidence
171	Nemotron 3 Super 100B	73.3%	estimated ± 3.2 pp, high confidence
172	Kimi K2	73.0%	estimated ± 3.2 pp, high confidence
173	Nemotron 3 Nano Omni 30B A3B	72.2%	measured
174	ZAYA1-8B	71.0%	measured
175	MiniCPM5-2B	70.2%	estimated ± 0.4 pp, high confidence
176	LongCat-Flash-Lite-Sparse	69.5%	estimated ± 0.4 pp, high confidence
177	GPT-OSS 120B	69.4%	estimated ± 3.2 pp, high confidence
178	o1-preview	68.6%	estimated ± 3.2 pp, high confidence
179	Grok 4.1 Fast	68.2%	estimated ± 3.2 pp, high confidence
180	Mistral Small 4	68.2%	estimated ± 3.2 pp, high confidence
181	Mistral Small 4 (Reasoning)	68.2%	estimated ± 3.2 pp, high confidence
182	GLM-4.5-Air	67.5%	estimated ± 3.2 pp, high confidence
183	Ling 3.0 Tiny	67.4%	estimated ± 3.2 pp, high confidence
184	Granite 4.2 30B	66.4%	measured
185	GPT-4.1	66.3%	measured
186	GPT-4.1 mini	64.2%	measured
187	Granite 4.2 8B	64.1%	measured
188	Trinity-Large-Preview	63.3%	estimated ± 0.4 pp, high confidence
189	Llama 4 Maverick	62.7%	estimated ± 3.2 pp, high confidence
190	North Mini Code	62.3%	estimated ± 3.2 pp, high confidence
191	Gemini 2.5 Flash	62.0%	estimated ± 3.2 pp, high confidence
192	DeepSeek V3 0324	61.4%	estimated ± 3.2 pp, high confidence
193	Claude 3.5 Sonnet	59.4%	measured
194	DeepSeek V3	59.1%	measured
195	Mistral Large 3	59.0%	estimated ± 3.2 pp, high confidence
196	Ling 2.6 Flash	59.0%	measured
197	Gemma 4 E4B	58.6%	measured
198	Mistral Medium 3	57.8%	estimated ± 3.2 pp, high confidence
199	Mellum2-12B-A2.5B-Thinking	57.6%	measured
200	GPT-OSS 20B	57.4%	estimated ± 3.2 pp, high confidence
201	ZAYA1-74B-Preview	57.3%	measured
202	Nemotron 3 Nano 30B	57.0%	estimated ± 3.2 pp, high confidence
203	Sarvam 105B	56.3%	estimated ± 3.2 pp, high confidence
204	Claude 3 Opus	55.9%	estimated ± 3.2 pp, high confidence
205	Granite 4.2 3B	54.8%	measured
206	GPT-4o	54.2%	estimated ± 3.2 pp, high confidence
207	LFM2.5-2.6B	53.9%	estimated ± 3.2 pp, high confidence
208	DeepSeek R1 Distill Qwen 32B	53.8%	estimated ± 3.2 pp, high confidence
209	Llama 4 Scout	51.9%	estimated ± 3.2 pp, high confidence
210	Gemini 1.5 Pro	50.7%	estimated ± 3.2 pp, high confidence
211	GPT-4.1 nano	50.3%	measured
212	Solar Pro 3	50.1%	estimated ± 3.2 pp, high confidence
213	Qwen3-Omni-30B-A3B-Thinking	49.7%	estimated ± 3.2 pp, medium confidence
214	Ultravox v0.6 Llama 3.3 70B	49.0%	estimated ± 3.2 pp, medium confidence
215	Mistral Large 2	48.3%	estimated ± 3.2 pp, medium confidence
216	Nemotron Ultra 253B	48.1%	estimated ± 3.2 pp, medium confidence
217	Llama 3.1 405B	46.4%	estimated ± 3.2 pp, medium confidence
218	LFM2.5-8B-A1B	45.8%	estimated ± 3.2 pp, medium confidence
219	GPT-4 Turbo	44.5%	estimated ± 3.2 pp, medium confidence
220	LLaDA2.2-mini	44.4%	estimated ± 0.4 pp, high confidence
221	Solar Pro 2	44.2%	estimated ± 3.2 pp, medium confidence
222	Nova Pro	43.9%	estimated ± 3.2 pp, medium confidence
223	Gemma 4 E2B	43.4%	measured
224	Soofi S 30B-A3B	43.4%	measured
225	Qwen2.5 Coder 32B Instruct	42.1%	estimated ± 3.2 pp, medium confidence
226	GPT-4o mini	41.5%	estimated ± 3.2 pp, medium confidence
227	Mellum2-12B-A2.5B-Instruct	40.9%	measured
228	Sarvam 30B	40.7%	estimated ± 3.2 pp, medium confidence
229	Celeris-1	39.0%	estimated ± 3.2 pp, medium confidence
230	Exaone 4.0 32B	38.8%	estimated ± 3.2 pp, medium confidence
231	Qwen3-Omni-30B-A3B-Instruct	36.1%	estimated ± 3.2 pp, medium confidence
232	Phi-4	35.4%	estimated ± 3.2 pp, medium confidence
233	Phi-4 Multimodal Instruct	34.4%	estimated ± 3.2 pp, medium confidence
234	Claude 3 Haiku	32.4%	estimated ± 3.2 pp, medium confidence
235	Gemini 1.0 Pro	30.3%	estimated ± 3.2 pp, medium confidence
236	Exaone 4.0 1.2B	29.3%	estimated ± 3.2 pp, medium confidence
237	Granite-4.0-H-1B	29.0%	estimated ± 3.2 pp, medium confidence
238	MiniCPM5-1B	26.3%	estimated ± 0.4 pp, high confidence
239	Gemma 3 27B	26.0%	estimated ± 3.2 pp, medium confidence
240	Granite-4.0-350M	25.8%	estimated ± 3.2 pp, medium confidence
241	Granite-4.0-H-350M	25.8%	estimated ± 3.2 pp, medium confidence
242	LFM2.5-VL-1.6B-Extract	25.8%	estimated ± 3.2 pp, medium confidence
243	LFM2.5-VL-450M	25.7%	measured
244	LFM2.5-230M	25.4%	measured

GPQA leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General