Knowledge & reasoning

AA-GPQA Diamond leaderboard

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

As of 2026-10-07, the highest measured score on AA-GPQA Diamond is 96.1% by GPT-6 Astra. 63 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 4 Argon	96.6%	estimated ± 1.2 pp, low confidence
2	Claude Sonnet 5.5	96.2%	estimated ± 2.1 pp, low confidence
3	Claude Opus 5.5	96.2%	estimated ± 2.1 pp, low confidence
4	Ling 3.1 Flash	96.2%	estimated ± 2.1 pp, low confidence
5	GPT-6 Astra	96.1%	measured
6	Gemini 3.8 Flash	95.3%	measured
7	Grok 4.6	94.9%	measured
8	Gemini 3.7 Flash	94.5%	measured
9	GPT-6.1 Sol	94.4%	estimated ± 1.2 pp, high confidence
10	GPT-6 Sol	94.3%	estimated ± 1.2 pp, high confidence
11	Gemini 3.1 Pro	94.1%	measured
12	GPT-5.6 Sol	94.1%	measured
13	Claude Fable 5.1	93.7%	measured
14	Claude Mythos 5	93.6%	estimated ± 2.7 pp, high confidence
15	GPT-5.5 Pro	93.5%	estimated ± 2.7 pp, high confidence
16	GPT-5.5	93.5%	measured
17	Kimi K3	93.5%	measured
18	Muse Spark 1.3	93.5%	measured
19	GPT-5.4 Pro	93.4%	estimated ± 2.7 pp, high confidence
20	dots3-note Preview	93.2%	estimated ± 2.7 pp, high confidence
21	Claude Opus 5	93.2%	measured
22	Grok 4.5	93.1%	measured
23	MiniMax M3	92.9%	measured
24	DeepSeek V4 Pro 0813	92.8%	measured
25	Gemini 3.6 Flash	92.8%	measured
26	Qwen3.8 Max Preview	92.8%	measured
27	Claude Fable 5	92.6%	measured
28	GPT-5.6 Terra	92.5%	measured
29	Qwen3.7 Max	92.3%	measured
30	Qwen3.8-Flash-Next	92.3%	measured
31	Gemini 3.5 Flash	92.2%	measured
32	Sakana Fugu-Ultra	92.2%	estimated ± 2.7 pp, high confidence
33	Pareto 26.9	92.0%	estimated ± 2.7 pp, high confidence
34	Claude Opus 4.8	92.0%	measured
35	GPT-5.4	92.0%	measured
36	Pareto 26.10 Preview	92.0%	estimated ± 5.6 pp, medium confidence
37	MiMo-V2.6-Pro	91.8%	estimated ± 6.8 pp, medium confidence
38	GLM-5.3	91.7%	measured
39	Sakana Fugu	91.7%	estimated ± 2.7 pp, high confidence
40	GPT-5.3 Codex	91.5%	measured
41	Grok 4.7	91.5%	estimated ± 2.1 pp, medium confidence
42	Claude Haiku 5.5	91.4%	estimated ± 2.7 pp, high confidence
43	Claude Opus 4.7 (Adaptive)	91.4%	measured
44	GLM-5.3-Flash	91.2%	measured
45	Ornith-1.5-397B	91.2%	estimated ± 2.7 pp, high confidence
46	Claude Sonnet 5	91.1%	measured
47	GPT-5.6 Luna	91.1%	measured
48	Kimi K2.6	91.1%	measured
49	Qwen3.8 Max	91.0%	estimated ± 2.7 pp, high confidence
50	Hy4 preview	90.9%	estimated ± 2.7 pp, high confidence
51	DeepSeek V4 Flash 0731	90.8%	measured
52	Gemini 3 Pro	90.8%	measured
53	Mistral Large 4	90.8%	estimated ± 6.8 pp, medium confidence
54	MiMo-V2.6-Flash	90.7%	estimated ± 6.8 pp, medium confidence
55	Qwen3.8-27B	90.5%	measured
56	Muse Spark 1.2	90.4%	measured
57	GPT-5.2	90.3%	measured
58	Grok 4.3	90.1%	measured
59	Qwen3.7 Plus	90.0%	measured
60	GPT-5.2-Codex	89.9%	measured
61	Muse Spark 1.1	89.8%	measured
62	Hy3	89.7%	measured
63	Hy3 Preview	89.7%	measured
64	Claude Opus 4.6 (Adaptive)	89.6%	measured
65	Kimi K2.7 Code	89.6%	measured
66	GLM-5.2	89.5%	measured
67	Inkling-Small	89.5%	measured
68	Agents-A1	89.3%	estimated ± 3.4 pp, high confidence
69	Step 5 Preview	89.3%	estimated ± 3.4 pp, high confidence
70	Beam	89.3%	estimated ± 2.7 pp, high confidence
71	GPT-6 Luna	89.3%	estimated ± 1.2 pp, medium confidence
72	Solar Pro 4	89.1%	measured
73	Gemini 3 Pro Deep Think	89.1%	estimated ± 2.7 pp, high confidence
74	Qwen 3.6 Max (preview)	88.8%	measured
75	Grok 4.20	88.7%	estimated ± 1.2 pp, medium confidence
76	DeepSeek V4.1 Flash	88.7%	estimated ± 3.4 pp, high confidence
77	Qwen3.8-Omni-Flash	88.7%	estimated ± 3.4 pp, high confidence
78	Claude Opus 4.7	88.5%	measured
79	Interfaze Beta	88.5%	estimated ± 5.4 pp, medium confidence
80	Muse Spark	88.4%	measured
81	Qwen3.6 Plus	88.2%	measured
82	Kimi K2.5	87.9%	measured
83	Kimi K2.5 (Reasoning)	87.9%	measured
84	Grok 4	87.7%	measured
85	GPT-5.4 mini	87.5%	measured
86	MiniMax M2.7	87.4%	measured
87	GPT-5.1	87.3%	measured
88	Inkling	87.2%	measured
89	Solar Open 2	87.2%	estimated ± 2.7 pp, high confidence
90	MiMo-V2-Pro	87.0%	measured
91	GLM-5.1	86.8%	measured
92	Nemotron 3 Ultra	86.7%	measured
93	Claude Opus 4.5 Thinking	86.6%	measured
94	MiMo-V2.5-Pro	86.6%	measured
95	Apodex 1.1	86.4%	measured
96	Apodex 1.1 Mini	86.4%	measured
97	Ling 3.0 Flash VL	86.2%	measured
98	Qwen3.5 397B	86.1%	measured
99	Qwen3.5 397B (Reasoning)	86.1%	measured
100	Seed 2.1 Pro	86.1%	estimated ± 5.0 pp, medium confidence
101	Ornith-1.5-35B-A3B	86.1%	estimated ± 2.7 pp, high confidence
102	GPT-5.1-Codex	86.0%	measured
103	GPT-5.1-Codex-Max	86.0%	measured
104	GLM-4.7	85.9%	measured
105	Qwen3.5-27B	85.8%	measured
106	A.X K2	85.7%	measured
107	Gemma 4 31B	85.7%	measured
108	Qwen3.5-122B-A10B	85.7%	measured
109	Seed 2.1 Turbo	85.6%	estimated ± 5.0 pp, medium confidence
110	Ling 3.0 Flash	85.5%	measured
111	Ling 3.0 Flash FP8	85.5%	measured
112	Claude Sonnet 4.5 Thinking	85.5%	estimated ± 2.7 pp, high confidence
113	Claude Sonnet 4.5	85.5%	estimated ± 2.7 pp, high confidence
114	GPT-5 (high)	85.4%	measured
115	Grok 4.1 Fast (Reasoning)	85.3%	measured
116	GLM-5-Turbo	84.7%	measured
117	Grok 4 Fast (Reasoning)	84.7%	measured
118	o3-pro	84.5%	measured
119	Qwen3.5-35B-A3B	84.5%	measured
120	Gemini 2.5 Pro	84.4%	measured
121	Ternary Bonsai 2 27B	84.3%	estimated ± 5.4 pp, medium confidence
122	GPT-5 (medium)	84.2%	measured
123	Qwen3.6-27B	84.2%	measured
124	Qwen3.6-35B-A3B	84.1%	measured
125	Qwen3.5 Flash	84.1%	estimated ± 4.2 pp, high confidence
126	Claude Opus 4.6	84.0%	measured
127	Ornith-1.5-9B	83.9%	estimated ± 2.7 pp, high confidence
128	Gemini 3.5 Flash-Lite	83.8%	measured
129	Muse Glimmer 30B	83.5%	measured
130	LongCat-Flash-Lite-Sparse	83.0%	estimated ± 2.0 pp, low confidence
131	K-EXAONE 2.0	82.9%	measured
132	Qwen3 235B 2507	82.8%	estimated ± 5.0 pp, medium confidence
133	MiMo-V2-Omni	82.8%	measured
134	MAI-Thinking-1	82.7%	estimated ± 5.4 pp, medium confidence
135	o3	82.7%	measured
136	MiMo-V2.5	82.5%	estimated ± 4.2 pp, high confidence
137	Claude 4.1 Opus	82.4%	estimated ± 6.8 pp, medium confidence
138	GLM-5	82.0%	measured
139	Gemini 3.1 Flash-Lite	81.9%	estimated ± 4.2 pp, high confidence
140	GPT-5.4 nano	81.7%	measured
141	DeepSeek-R1	81.3%	measured
142	Gemini 3 Flash	81.2%	measured
143	Claude Opus 4.5	81.0%	measured
144	Claude 4.1 Opus Thinking	80.9%	measured
145	GLM-5V-Turbo	80.9%	measured
146	Step 3.7 Flash	80.9%	measured
147	Nemotron 3 Super 100B	80.0%	measured
148	Claude Sonnet 4.6	79.9%	measured
149	Gemma 4 26B A4B	79.2%	measured
150	K-Exaone	78.3%	measured
151	GPT-OSS 120B	78.2%	measured
152	Mercury 2.5	78.0%	estimated ± 5.6 pp, medium confidence
153	DeepSeek V3.1 (Reasoning)	77.9%	measured
154	o1-pro	77.5%	estimated ± 5.4 pp, medium confidence
155	Mistral Small 4	76.9%	measured
156	Mistral Small 4 (Reasoning)	76.9%	measured
157	Kimi K2	76.6%	measured
158	o1-preview	76.5%	measured
159	Qwen3 Max	76.4%	measured
160	Command A+	76.1%	measured
161	Nemotron 3 Nano 30B	75.7%	measured
162	North Mini Code	75.7%	measured
163	Gemma 4 12B	75.3%	measured
164	Trinity-Large-Preview	75.2%	measured
165	Trinity-Large-Thinking	75.2%	measured
166	DeepSeek V3.2	75.1%	measured
167	Mistral Medium 3.5 128B	74.8%	measured
168	o3-mini	74.8%	measured
169	o1	74.7%	measured
170	Nemotron 3.5 Lightning 30B A3B NVFP4	74.3%	measured
171	Sarvam 105B	73.8%	measured
172	DeepSeek V3.1	73.5%	measured
173	Ling 3.0 Tiny	73.4%	measured
174	GLM-4.5-Air	73.3%	measured
175	Quasar 438B	73.2%	measured
176	Nemotron Ultra 253B	72.8%	measured
177	Grok Code Fast 1	72.7%	measured
178	Qwen3-Omni-30B-A3B-Thinking	72.6%	measured
179	Claude Haiku 4.5	72.5%	estimated ± 4.2 pp, high confidence
180	GLM-4.5	72.5%	estimated ± 4.2 pp, high confidence
181	Solar Pro 3	72.4%	measured
182	MiniCPM5-2B	70.2%	measured
183	MiniCPM5-1B	70.2%	estimated ± 5.0 pp, low confidence
184	ZAYA1-8B	69.4%	estimated ± 5.4 pp, medium confidence
185	Laguna XS.2	68.9%	estimated ± 4.2 pp, high confidence
186	GPT-OSS 20B	68.8%	measured
187	Laguna M.1	68.7%	estimated ± 4.2 pp, medium confidence
188	Claude 4 Sonnet	68.3%	measured
189	Gemini 2.5 Flash	68.3%	measured
190	Mistral Large 3	68.0%	measured
191	Llama 4 Maverick	67.1%	measured
192	GPT-4.1	66.6%	measured
193	GPT-4.1 mini	66.4%	measured
194	MiMo-V2-Flash	65.6%	measured
195	DeepSeek V3 0324	65.5%	measured
196	Granite 4.2 30B	64.4%	measured
197	Grok 4.1 Fast	63.7%	measured
198	Sarvam 30B	63.3%	measured
199	GLM-4.6	63.2%	measured
200	Celeris-1	63.1%	measured
201	Granite 4.2 8B	63.1%	measured
202	Exaone 4.0 32B	62.8%	measured
203	Qwen3-Omni-30B-A3B-Instruct	62.0%	measured
204	DeepSeek R1 Distill Qwen 32B	61.5%	measured
205	Ling 2.6 Flash	59.3%	measured
206	Gemini 1.5 Pro	58.9%	measured
207	Llama 4 Scout	58.7%	measured
208	Mistral Medium 3	57.8%	measured
209	Claude 3.5 Sonnet	57.7%	estimated ± 5.4 pp, medium confidence
210	Gemma 4 E4B	57.6%	measured
211	Phi-4	57.5%	measured
212	Solar Pro 2	56.1%	measured
213	Granite 4.2 3B	55.9%	measured
214	Mellum2-12B-A2.5B-Thinking	55.9%	estimated ± 5.4 pp, medium confidence
215	LFM2.5-2.6B	55.8%	measured
216	DeepSeek V3	55.7%	measured
217	ZAYA1-74B-Preview	55.6%	estimated ± 5.4 pp, medium confidence
218	GPT-4o	54.3%	measured
219	GPT-4 Turbo	52.4%	estimated ± 6.8 pp, medium confidence
220	Llama 3.1 405B	51.5%	measured
221	LFM2.5-8B-A1B	51.3%	measured
222	GPT-4.1 nano	51.2%	measured
223	Nova Pro	49.9%	measured
224	Ultravox v0.6 Llama 3.3 70B	49.8%	measured
225	Claude 3 Opus	48.9%	measured
226	Mistral Large 2	48.6%	measured
227	Nemotron 3 Nano Omni 30B A3B	46.9%	measured
228	Gemma 4 E2B	43.3%	measured
229	Gemma 3 27B	42.8%	measured
230	GPT-4o mini	42.6%	measured
231	Exaone 4.0 1.2B	42.4%	measured
232	LLaDA2.2-mini	41.8%	estimated ± 5.6 pp, low confidence
233	Qwen2.5 Coder 32B Instruct	41.7%	measured
234	Soofi S 30B-A3B	41.6%	estimated ± 5.4 pp, medium confidence
235	Mellum2-12B-A2.5B-Instruct	39.1%	estimated ± 5.4 pp, low confidence
236	Claude 3 Haiku	37.4%	measured
237	Phi-4 Multimodal Instruct	31.5%	measured
238	LFM2.5-VL-1.6B-Extract	28.9%	measured
239	Gemini 1.0 Pro	27.7%	measured
240	Granite-4.0-H-1B	26.3%	measured
241	Granite-4.0-350M	26.1%	measured
242	Granite-4.0-H-350M	25.7%	measured
243	LFM2.5-VL-450M	23.7%	estimated ± 5.4 pp, low confidence
244	LFM2.5-230M	23.4%	estimated ± 5.4 pp, low confidence

AA-GPQA Diamond leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General