Which model leads on GPQA diamond?

Gemini 4 Carbon from Google DeepMind leads GPQA diamond with a score of 99.3. The median score across 184 tested models is 72.1.

Is GPQA diamond saturated?

Yes · the top model on GPQA diamond has reached 99.3 out of 100, within 5% of the theoretical ceiling. This benchmark is approaching saturation and may be replaced by a harder successor.

Does GPQA diamond predict performance on other benchmarks?

Yes · GPQA diamond scores correlate 0.98 with OpenCompass · MMLU-Pro across 15 shared models. Models that do well on GPQA diamond tend to do well on OpenCompass · MMLU-Pro.

How often is GPQA diamond data refreshed?

BenchGecko pulls updates daily. New model scores on GPQA diamond appear as soon as they are published by Epoch AI or the model provider.

Benchmark · ReasoningSettled

GPQA diamond

Name: GPQA diamond Benchmark
Creator: BenchGecko
License: https://creativecommons.org/licenses/by/4.0/

Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.

Updated 2026-11-10

The benchmark is approaching saturation.

Scoring: Exact-match accuracy on 198 multiple-choice questions. Four options per question. No partial credit.

Models tested

186

Top score

99.3

Gemini 4 Carbon

Median

72.1

min 1.4

Top-5 spread

σ 0.6

Settled

The Frontier

Best score over time · one chart, every benchmark

Chart type

Frontier on GPQA diamond rose from 71.9 to 99.3 in 23 months · +27.4 points · latest leader Gemini 4 Carbon from Google DeepMind.

Pink dots = frontier records · 13 totalClick to open model page

Full rankings

186 models tested · sorted by score · includes 7 verified scores

#	Model	Score	Price	Source
1	Gemini 4 Carbon· Google DeepMind	99.3	$2.00
2	Gemini 4 Argon· Google DeepMind	99.2	$2.00
3	GPT-6 Astra· OpenAI	98.3	$10.00
4	Claude Sonnet 5.5· Anthropic	98.0	$2.00
5	GPT-6.1 Sol· OpenAI	97.8	$2.00
6	Gemini 3.8 Flash· Google DeepMind	97.8	$0.82
7	Gemini 3.7 Flash· Google DeepMind	97.0	$0.75
8	Gemini 3.1 Pro Preview· Google DeepMind	96.5	$2.00
9	GPT-6 Sol· OpenAI	96.3	$2.00
10	GPT-6.1 Terra· OpenAI	96.3	$1.00
11	Kimi K3.1 Pro· moonshotai	96.1	$2.00
12	Gemini 3.6 Flash· Google DeepMind	96.0	$0.75
13	GPT-5.5· OpenAI	95.8	$5.00	OpenAI GPT-5.5 Blog, Apr 2026
14	Grok 4.6· xAI	95.8	$2.00
15	GPT-5.5 Pro· OpenAI	95.7	$30.00	OpenAI GPT-5.5 Blog, Apr 2026
16	Claude Opus 5· Anthropic	95.7	$5.00
17	Claude Fable 5.2· Anthropic	95.7	$8.00
18	GPT-5.6 Sol· OpenAI	95.1	$4.40
19	Grok 4.5· xAI	95.0	$2.00
20	GPT-5.6 Terra· OpenAI	94.9	$2.00
21	GPT-5.4· OpenAI	94.9	$2.50	OpenAI GPT-5.4 Blog, Mar 2026
22	DeepSeek V4.1 Flash 1017· DeepSeek	94.7	$0.30
23	Kimi K3· moonshotai	94.6	$0.19
24	Claude Mythos Preview· Anthropicverified	94.5	n/a	Anthropic Mythos System Card, Apr 2026
25	Gemini 3.1 Pro· Googleverified	94.3	n/a	Google Gemini 3.1 Pro Announcement, Mar 2026
26	Gemini 3.5 Flash· Google DeepMind	94.2	$1.50
27	Grok 4.7· xAI	94.1	$2.00
28	Qwen3.8 Max (0902)· Alibaba Qwen	94.0	$2.00
29	Gemini 3 Pro· Google DeepMind	93.9	n/a
30	GLM 5.2· z-ai	92.9	$0.01
31	GPT-5.4 Pro· OpenAI	92.8	$30.00
32	DeepSeek V4 Pro 0813· DeepSeek	92.6	$0.11
33	GPT-5.6 Luna· OpenAI	92.5	$0.20
34	GPT-5.2· OpenAI	92.2	$1.75
35	Claude Opus 4.8· Anthropic	91.7	$5.00
36	DeepSeek V4 Flash 0731· DeepSeek	91.7	$0.04
37	GLM 5.3· z-ai	91.5	$0.07
38	DeepSeek V4 Pro· DeepSeek	91.5	$0.96
39	MiniMax M3· minimax	91.5	$0.30
40	Qwen3.7 Max· Alibaba Qwen	91.5	$1.48
41	Kimi K2.6· moonshotai	91.4	$0.65
42	Claude Opus 5.5· Anthropic	91.1	$4.00
43	Claude Sonnet 5· Anthropic	91.0	$2.00
44	Claude Opus 4.6· Anthropic	91.0	$5.00	Anthropic Opus 4.6 System Card, Feb 2026
45	GPT-6 Luna· OpenAI	90.9	$0.10
46	Claude Opus 4.7· Anthropic	90.5	$5.00	Anthropic Opus 4.7 Announcement, Apr 2026
47	Mistral Large 4· Mistral AI	90.5	$0.70
48	GLM 5.3 Flash· z-ai	90.5	$0.04
49	GLM 5.1· z-ai	90.1	$1.40
50	DeepSeek V4.1 Flash· DeepSeek	90.0	$0.02
51	Muse Spark· Meta	90.0	n/a
52	Claude Haiku 5.5· Anthropic	89.8	$0.10
53	Gemini 3 Flash Preview· Google DeepMind	89.4	$0.50
54	Grok 4.20· xAI	89.3	$1.25
55	Grok 4.3· xAI	88.7	$1.25
56	Muse Spark Open· Meta	88.3	$1.25
57	Qwen3.6 Plus· Alibaba Qwen	88.0	$0.33
58	Qwen3.7 Plus· Alibaba Qwen	87.3	$0.32
59	Kimi K2.7 Code· moonshotai	87.3	$0.60
60	GPT-5.1· OpenAI	87.0	$1.25
61	Kimi K2.5· moonshotai	86.9	$0.53
62	Claude Sonnet 4.6· Anthropic	86.6	$3.00
63	Grok 4· xAI	86.1	n/a
64	GPT-5.4 Mini· OpenAI	85.9	$0.75
65	Qwen3.5 397B A17B· Alibaba Qwen	85.2	$0.60
66	GPT-5· OpenAI	85.0	$1.25
67	Claude Fable 5· Anthropic	84.5	$10.00
68	Qwen3.6 27B· Alibaba Qwen	84.5	$0.32
69	Nemotron 3 Ultra· NVIDIA	83.8	$0.60
70	GLM 5· z-ai	83.8	$1.00
71	Gemini 2.5 Pro· Google DeepMind	83.7	$1.25
72	Qwen3.6 Max Preview· Alibaba Qwen	83.2	$1.03
73	Qwen3.6 35B A3B· Alibaba Qwen	83.1	$0.10
74	Qwen 3.5 Plus (hosted 397B-A17B)· Alibaba Qwen	83.1	n/a
75	Claude Opus 4.5· Anthropic	81.4	$5.00
76	Qwen3.5-35B-A3B· Alibaba Qwen	81.2	$0.14
77	DeepSeek V3.2· DeepSeek	81.1	$0.26
78	Qwen3.6 Flash· Alibaba Qwen	81.0	$0.19
79	GLM 4.7· z-ai	81.0	$0.60
80	Gemini 3.5 Flash Lite· Google DeepMind	81.0	$0.30
81	Qwen3.7 Flash· Alibaba Qwen	79.6	$0.03
82	Claude Sonnet 4.5· Anthropic	79.6	$3.00
83	Qwen3.5-Flash· Alibaba Qwen	79.6	$0.07
84	Kimi K2 Thinking· moonshotai	79.0	$0.60
85	Gemini 3.1 Flash Lite· Google DeepMind	78.9	$0.25
86	Claude 3.7 Sonnet· Anthropic	76.0	n/a
87	o4 Mini· OpenAI	75.8	$1.10
88	o4 Mini (batch)· OpenAI	75.8	$0.55
89	o3· OpenAI	75.8	$2.00
90	Claude Sonnet 4· Anthropic	75.3	$3.00
91	Qwen3.5-9B· Alibaba Qwen	75.0	$0.07
92	GPT-5.4 Nano· OpenAI	74.3	$0.20
93	Qwen3 235B A22B Thinking 2507· Alibaba Qwen	73.4	$0.45
94	o3 Mini (batch)· OpenAI	72.3	$0.55
95	o1· OpenAI	71.9	$15.00
96	Grok-3 mini· xAI	71.2	n/a
97	Gemma 4 31B· Google DeepMind	70.5	$0.20
98	Claude Opus 4.1· Anthropic	69.7	$15.00
99	GPT-5 Mini· OpenAI	69.4	$0.25
100	o3 Mini· OpenAI	69.4	$1.10
101	R1 0528· DeepSeek	68.4	$0.50
102	Claude Opus 4· Anthropic	68.3	$15.00
103	gpt-oss-120b· OpenAI	67.7	$0.15
104	Grok 3· xAI	67.7	n/a
105	Gemma 4 26B A4B · Google DeepMind	67.0	$0.08
106	Qwen3 Max· Alibaba Qwen	63.5	$0.78
107	R1· DeepSeek	62.3	$0.70
108	GPT-5 Nano· OpenAI	61.7	$0.05
109	Claude Haiku 4.5· Anthropic	61.6	$1.00
110	Qwen3 235B A22B· Alibaba Qwen	60.9	$0.46
111	GPT-4.5· OpenAI	60.7	n/a
112	GPT-4.1 Mini· OpenAI	56.7	$0.40
113	Qwen3 32B· Alibaba Qwen	56.6	$0.08
114	Gemini 2.0 Pro· Google DeepMind	56.5	n/a
115	QwQ 32B· Alibaba Qwen	56.0	$0.15
116	Llama 4 Maverick· Meta	56.0	$0.35
117	GPT-4.1· OpenAI	55.9	$2.00
118	Gemini 2.0 Flash (Dec 2024)· Google DeepMind	54.4	n/a
119	Qwen3 14B· Alibaba Qwen	53.8	$0.10
120	DeepSeek R1 Distill Qwen 32B· DeepSeek	52.2	n/a
121	Gemini 2.0 Flash· Google DeepMind	52.2	$0.10
122	o1-mini· OpenAI	51.9	n/a
123	Qwen3 30B A3B· Alibaba Qwen	50.9	$0.12
124	gpt-oss-20b· OpenAI	49.7	$0.03
125	Mistral Medium 3· Mistral AI	46.0	$0.40
126	Gemini 2.0 Flash Thinking (Jan 2025)· Google DeepMind	44.5	n/a
127	DeepSeek V3· DeepSeek	43.8	$0.32
128	Qwen2.5-Max· Alibaba Qwen	43.2	n/a
129	Qwen3 8B· Alibaba Qwen	42.3	$0.12
130	Phi 4· Microsoft	41.4	$0.07
131	Grok-2 (Dec 2024)· xAI	40.0	n/a
132	Claude 3.5 Sonnet· Anthropic	38.7	n/a
133	Llama 3.1 405B· Meta	36.0	n/a
134	Llama 4 Scout· Meta	35.8	$0.10
135	o1-preview· OpenAI	35.2	n/a
136	Mistral Large 2411· Mistral AI	35.1	$2.00
137	Qwen2.5 72B Instruct· Alibaba Qwen	33.5	$0.36
138	GPT-4.1 Nano· OpenAI	33.2	$0.10
139	GPT-4o (2024-08-06)· OpenAI	32.3	$2.50
140	Mistral Small 3.2 24B· Mistral AI	32.1	$0.10
141	Mistral Large 2407· Mistral AI	32.0	$2.00
142	GPT-4o (2024-05-13)· OpenAI	31.9	$5.00
143	Gemma 3 27B· Google DeepMind	31.8	$0.08
144	Gemma 3 27B (free)· Google DeepMind	31.8	$0.00
145	U Magistral Small 1.1· Unknown	31.2	n/a
146	Mistral Small 3.1 24B· Mistral AI	31.2	$0.35
147	Llama 3.3 70B Instruct· Meta	31.2	$0.71
148	Claude 3 Opus· Anthropic	30.8	n/a
149	GPT-4o (2024-11-20)· OpenAI	30.5	$2.50
150	Llama 3.3 70B Instruct (free)· Meta	29.9	$0.00
151	Mistral Small 3· Mistral AI	29.7	$0.05
152	Gemini 1.5 Pro (May 2024)· Google DeepMind	29.0	n/a
153	Qwen2.5 32B Instruct· Alibaba	28.1	n/a
154	Gemini 1.5 Pro (Feb 2024)· Google DeepMind	27.8	n/a
155	Llama 3.1 70B Instruct· Meta	26.7	$0.72
156	DeepSeek R1 Distill Qwen 14B· DeepSeek	26.3	n/a
157	Llama 3.2 90B· Meta	22.3	n/a
158	Qwen2-72B· Alibaba Qwen	21.9	n/a
159	Claude 3 Sonnet· Anthropic	21.7	n/a
160	Llama 3-70B· Meta	21.6	n/a
161	Gemini 1.5 Flash (May 2024)· Google DeepMind	21.4	n/a
162	Llama 3 70B Instruct· Meta	20.8	$0.51
163	Gemma 3 12B· Google DeepMind	20.1	$0.05
164	Mistral Large· Mistral AI	19.1	$2.00
165	GPT-4o-mini· OpenAI	17.7	$0.15
166	Claude 3.5 Haiku· Anthropic	17.5	n/a
167	GPT-4o-mini (2024-07-18)· OpenAI	17.0	$0.15
168	Gemma 2 27B· Google DeepMind	16.0	$0.65
169	Claude 3 Haiku· Anthropic	15.7	$0.25
170	Qwen2.5 7B Instruct· Alibaba Qwen	14.6	$0.10
171	GPT-4 (older v0314)· OpenAI	14.3	$30.00
172	Claude 2· Anthropic	13.4	n/a
173	Mixtral 8x22B Instruct· Mistral AI	12.6	$2.00
174	Gemini 1.0 Pro· Google DeepMind	12.4	n/a
175	Claude 2.1· Anthropic	11.1	n/a
176	GPT-4 Turbo (batch)· OpenAI	7.8	$5.00
177	GPT-4 Turbo· OpenAI	7.8	$10.00
178	Mixtral 8x7B· Mistral AI	7.8	n/a
179	Mixtral 8x7B Instruct· Mistral AI	7.5	$0.54
180	Mistral Nemo· Mistral AI	6.8	$0.03
181	phi-3-medium 14B· Microsoft	3.6	n/a
182	Gemma 2 9B· Google DeepMind	3.3	$0.03
183	GPT-3.5 Turbo (older v0613)· OpenAI	3.0	$1.00
184	Llama 3.1 8B Instruct· Meta	2.7	$0.02
185	Llama 2-13B· Meta	1.8	n/a
186	Llama 3 8B Instruct· Meta	1.4	$0.14

Details

Category: Reasoning
Creator: NYU (Rein et al.)
Max score: 100
Dataset: 198 questions
Modality: Text
Format: JSON
License: CC-BY-4.0
Scoring: Exact-match accuracy on 198 multiple-choice questions. Four options per question. No partial credit.
Models: 186
Published: 2023-11-20
Updated: 2026-11-10

Tests

Expert-level sciencePhysicsBiologyChemistryReasoning under uncertainty

Does not test

CodeLong contextSpeedTool useVisionCreative writing

Links

GPQA Paper (Rein et al., 2023)Anthropic Opus 4.7 Announcement OpenAI GPT-5.4 Blog Learn more

Gecko's Take

“GPQA Diamond served its purpose brilliantly, but the ceiling is cracking. When models outscore the PhDs who wrote the questions, we need GPQA Platinum.”

Related benchmarks