Archived Benchmark

Since performance on this benchmark has saturated, we no longer run this benchmark on new model releases. Previous results are preserved here for posterity.

Academic

GPQA Diamond

Updated 9/1/2026

Graduate-level Google-Proof Q&A benchmark evaluating models on questions that require deep reasoning.

As of September 1, 2026, Gemini 3.1 Pro Preview (02/26) ranks first on GPQA Diamond with 95.45%, followed by GPT-5.6 Sol (95.20%) and Grok 4.6 (94.70%).

GPQA DiamondGraduate Google-proof reasoning Q&A
ACCURACY

GPQA Diamond leaderboard

Rank Model Accuracy Cost In / Out Latency
1 Gemini 3.1 Pro Preview (02/26) 95.45% $2 / $12 65.76s
2 GPT-5.6 Sol 95.20% $4 / $20 58.30s
3 Grok 4.6 94.70% $2 / $6 3m29s
4 Gemini 3.8 Flash 94.44% $1.5 / $7.5 20.18s
5 Gemini 3.7 Flash 93.94% $1.5 / $7.5 8.77s
6 Qwen 3.8 Max 93.69% $2 / $6 4m27s
7 Gemini 3.6 Flash 93.43% $1.5 / $7.5 18.06s
8 Claude Opus 5 93.43% $5 / $25 39.59s
9 Claude Fable 5.1 93.43% $10 / $50 45.86s
10 Claude Fable 5 93.18% $10 / $50 99.90s
11 GPT 5.5 93.18% $5 / $30 112.33s
12 Kimi K3 92.93% $3 / $15 2m02s
13 Grok 4.5 92.93% $2 / $6 3m38s
14 Gemini 3.5 Flash 92.68% $1.5 / $9 27.66s
15 MiniMax-M3 92.68% $0.6 / $2.4 4m39s
16 Claude Opus 4.8 92.42% $5 / $25 2m14s
17 DeepSeek V4 Pro 0813 92.42% $1.32 / $3.96 4m39s
18 GPT-5.6 Luna 91.67% $0.2 / $1.2 53.52s
19 Gemini 3 Pro (11/25) 91.67% $2 / $12 61.34s
20 GPT 5.2 91.67% $1.75 / $14 2m53s
21 GPT 5.4 (xhigh) 91.67% $2.5 / $15 3m23s
22 Grok 4.3 91.41% $1.25 / $2.5 12m20s
23 Muse Spark 1.1 91.16% $1.25 / $4.25 75.20s
24 GPT-5.6 Terra 90.91% $2 / $12 36.76s
25 Qwen 3.7 Max 90.15% $2.5 / $7.5 2m07s
26 Claude Opus 4.7 90.15% $5 / $25 3m23s
27 DeepSeek V4 Flash 0731 89.90% $0.44 / $1.32 79.96s
28 Muse Spark 89.65% N/A 94.58s
29 Claude Opus 4.6 (Thinking) 89.65% $5 / $25 5m19s
30 DeepSeek V4 89.39% $1.32 / $3.96 15m00s
31 Kimi K2.6 89.14% $0.95 / $4 9m05s
32 Claude Sonnet 5 88.89% $2 / $10 63.23s
33 Qwen 3.8 27B 88.89% $0.5 / $3 116.36s
34 Grok 4.20 (Reasoning) 88.64% $2 / $6 59.42s
35 Grok 4 88.13% $3 / $15 115.52s
36 GLM 5.3 88.13% $1.4 / $4.4 3m31s
37 Gemini 3 Flash (12/25) 87.88% $0.5 / $3 83.84s
38 Qwen 3.5 Plus 87.37% $0.4 / $2.4 103.77s
39 Qwen 3.6 Plus 87.37% $0.5 / $3 2m02s
40 Inkling 87.12% $1 / $4.05 5m09s
41 GPT 5.1 86.62% $1.25 / $10 85.71s
42 MiniMax-M2.7 86.62% $0.3 / $1.2 2m52s
43 GLM 5.3 Flash 86.36% $0.075 / $0.25 2m11s
44 Nemotron 3 Ultra 86.11% N/A 2m38s
45 Claude Opus 4.5 (Thinking) 85.86% $5 / $25 2m19s
46 GPT 5 85.61% $1.25 / $10 2m50s
47 GLM 5.2 85.61% $1.4 / $4.4 3m45s
48 Claude Sonnet 4.6 85.61% $3 / $15 6m18s
49 Grok 4 Fast (Reasoning) 85.35% $0.2 / $0.5 21.41s
50 Ling 3.0 Flash 84.85% $0.075 / $0.22 43.51s
51 Qwen 3 Max Thinking 84.85% $1.2 / $6 6m08s
52 GLM 5.1 84.52% $1 / $3.2 4m10s
53 Grok 4.1 Fast (Reasoning) 84.34% $0.2 / $0.5 68.48s
54 o3 84.09% $2 / $8 65.78s
55 Kimi K2.5 84.09% $0.6 / $3 2m32s
56 Gemini 3.5 Flash Lite 83.84% $0.3 / $2.5 13.85s
57 Inkling Small 83.59% $0.3 / $1.2 4m20s
58 GLM 5 83.33% $1 / $3.2 4m03s
59 GPT 5.4 Mini 83.08% $0.75 / $4.5 63.11s
60 Qwen 3.5 Flash 82.83% $0.1 / $0.4 79.88s
61 MiMo V2.5 Pro 82.58% $0.435 / $0.87 4m11s
62 MiniMax-M2.5 82.07% $0.3 / $1.2 2m52s
63 Claude Sonnet 4.5 (Thinking) 81.63% $3 / $15 117.25s
64 Gemini 2.5 Flash Preview (9/25) (Nonthinking) 81.57% $0.3 / $2.5 19.48s
65 MiMo V2.5 81.57% $0.14 / $0.28 89.39s
66 Gemini 3.1 Flash Lite Preview 81.06% $0.25 / $1.5 16.46s
67 Gemini 2.5 Pro Exp 80.81% $1.25 / $10 41.10s
68 GPT 5 Mini 80.30% $0.25 / $2 67.13s
69 DeepSeek V3.2 (Thinking) 80.30% $0.56 / $1.68 4m20s
70 GLM 4.7 80.05% $0.6 / $2.2 3m31s
71 Claude Opus 4.5 (Nonthinking) 79.55% $5 / $25 13.54s
72 Qwen 3 Max 79.55% $1.2 / $6 70.15s
73 Grok 3 Mini Reasoning 79.29% $0.3 / $0.5 40.62s
74 GPT OSS 120B 78.54% $0.15 / $0.6 56.18s
75 MiniMax-M2.1 78.54% $0.3 / $1.2 2m15s
76 Kimi K2 Thinking 78.54% $0.6 / $2.5 9m46s
77 Qwen 3 Max Preview 77.78% $1.2 / $6 106.78s
78 GPT 5.4 Nano 77.53% $0.2 / $1.25 29.50s
79 Gemini 2.5 Flash Preview (9/25) (Thinking) 76.52% $0.3 / $2.5 23.53s
80 Claude Opus 4.1 (Thinking) 76.26% $15 / $75 35.77s
81 DeepSeek V3.2 (Nonthinking) 76.26% $0.56 / $1.68 46.83s
82 o3 Mini 75.50% $1.1 / $4.4 99.98s
83 Claude 3.7 Sonnet (Thinking) 75.25% $3 / $15 2m35s
84 Claude Sonnet 4 (Thinking) 75.00% $3 / $15 118.46s
85 o4 Mini 74.50% $1.1 / $4.4 30.00s
86 GLM 4.6 74.50% $0.6 / $2.2 7m55s
87 Grok 3 74.24% $3 / $15 29.91s
88 o1 73.23% $15 / $60 40.18s
89 Grok 3 Mini Reasoning 72.98% $0.3 / $0.5 14.23s
90 Claude Haiku 4.5 (Thinking) 72.22% $1 / $5 83.20s
91 GLM 4.5 72.22% $0.6 / $2.2 11m44s
92 Claude Opus 4 (Nonthinking) 71.72% $15 / $75 19.13s
93 Kimi K2 Instruct 71.46% $1 / $3 41.43s
94 Gemini 2.5 Flash Lite (9/25) (Thinking) 70.20% $0.1 / $0.4 20.61s
95 Qwen 3 (235B) 70.20% $0.22 / $0.88 5m06s
96 Claude Opus 4.1 (Nonthinking) 69.95% $15 / $75 35.93s
97 Llama 4 Maverick 69.44% $0.22 / $0.88 10.07s
98 Claude Sonnet 4 (Nonthinking) 69.44% $3 / $15 12.82s
99 GPT OSS 20B 68.94% $0.07 / $0.3 2m25s
100 Mistral Large 3 68.43% $0.5 / $1.5 21.22s
101 GPT 4.1 Mini 67.93% $0.4 / $1.6 9.28s
102 Claude 3.7 Sonnet (Nonthinking) 67.42% $3 / $15 9.22s
103 GPT 4.1 65.40% $2 / $8 23.14s
104 Gemini 2.0 Flash (001) 65.15% $0.1 / $0.4 5.80s
105 Grok 4.1 Fast Non-Reasoning 65.15% $0.2 / $0.5 8.66s
106 Gemini 2.5 Flash Lite (9/25) (Nonthinking) 64.14% $0.1 / $0.4 7.27s
107 GPT 5 Nano 63.38% $0.05 / $0.4 49.72s
108 Llama 3.3 Nemotron Super (Thinking) 62.37% N/A 77.59s
109 Magistral Medium 1.2 (09/2025) 62.37% $2 / $5 94.14s
110 Grok 4 Fast (Non-Reasoning) 62.12% $0.2 / $0.5 3.67s
111 DeepSeek V3 (03/24/2025) 61.62% $0.9 / $0.9 23.48s
112 Claude 3.5 Sonnet Latest 59.34% $3 / $15 7.24s
113 MiMo V2 Flash 59.34% $0.1 / $0.3 20.90s
114 Gemini 1.5 Pro (002) 58.33% $1.25 / $5 7.71s
115 Magistral Small 1.2 (09/2025) 58.33% $0.5 / $1.5 46.25s
116 Gemini 2.5 Flash Preview 4/17 (Nonthinking) 57.58% $0.3 / $2.5 43.45s
117 Laguna XS.2 55.05% N/A 69.31s
118 DeepSeek V3 54.55% $0.9 / $0.9 21.28s
119 GPT 4o (2024-11-20) 53.79% $2.5 / $10 14.92s
120 GPT 4.1 Nano 50.76% $0.1 / $0.4 4.19s
121 Mistral Small (02/2024) 50.76% $0.2 / $0.6 7.99s
122 Grok 2 50.76% $2 / $10 20.18s
123 GPT 4o (2024-05-13) 50.25% $5 / $15 10.08s
124 Llama 3.3 Instruct Turbo (70B) 50.00% $0.88 / $0.88 7.34s
125 Command A 48.48% $2.5 / $10 11.49s
126 Mistral Large (11/2024) 47.73% $2 / $6 15.80s
127 Llama 3.3 Nemotron Super (Nonthinking) 47.73% N/A 16.48s
128 Llama 4 Scout 46.97% $0.18 / $0.59 12.09s
129 Gemini 1.5 Flash (002) 45.96% $0.075 / $0.3 3.39s
130 Mistral Small 3.1 (03/2025) 44.19% $0.075 / $0.3 8.82s
131 GPT 4o Mini 44.19% $0.15 / $0.6 15.32s
132 Claude 3.5 Haiku Latest 37.88% $0.8 / $4 6.56s
133 Jamba 1.6 Large 36.11% $2 / $8 15.47s
134 Mistral Medium 3.5 34.85% $1.5 / $7.5 4m16s
135 Jamba 1.6 Mini 32.07% $0.2 / $0.4 4.95s
136 Command R+ 31.06% $2.5 / $10 8.19s
137 GPT 3.5 30.56% $0.5 / $1.5 2.63s
138 Laguna M.1 27.02% N/A 5m28s

Results



Dataset and Context

The Graduate-Level Google-Proof Q&A (GPQA) benchmark (paper) is a public academic benchmark commonly used to measure models’ general question-answering performance. It evaluates language models on challenging, graduate-level questions across STEM fields. When created, the questions were specifically designed to be “Google-proof” - they require deep understanding and reasoning rather than fact recall or search.

As in the original paper, this benchmark uses two evaluation approaches:

  • Zero-shot chain-of-thought: Models are asked to solve the problems with instruction to explain their reasoning steps
  • Few-shot chain-of-thought: In addition to the instructions, models are provided with 5 example questions and answers.

We focus on the “diamond” subset of 198 questions - these were questions in which expert validators answered correctly and no more than one out of three non-experts answered correctly. They are both challenging, but have unambigous answers.


Additional Notes

Methodology

We only used the diamond subset (198 questions) as the evaluation set - these were the questions that were most challenging for non-experts, but had the highest agreement rate between experts.

We replicated the original paper’s prompting and response parsing techniques as closely as possible, although for zeros-shot CoT, we only prompt the model once, rather than asking the question across two separate queries. Here are the prompts used:

Zero-shot CoT Prompt:

What is the correct answer to this question:

Two quantum states with energies E1 and E2 have a lifetime of 10^-9 sec and 10^-8 sec, respectively. We want to clearly distinguish these two energy levels. Which one of the following options could be their energy difference so that they can be clearly resolved?

Choices:
(A) 10^-8 eV
(B) 10^-9 eV
(C) 10^-4 eV
(D) 10^-11 eV

Reason through your answer step-by-step. Then, based on your reasoning, provide the single most likely answer choice. Answer in the format "The correct answer is (insert answer here)."

Few-shot CoT Prompt:

Here are some example questions from experts. An explanation is given before the final answer.
Answer the final question yourself, giving your reasoning beforehand.

Question: In a given population, 1 out of every 400 people has a cancer caused by a completely recessive allele, b. Assuming the population is in Hardy-Weinberg equilibrium, which of the following is the expected proportion of individuals who carry the b allele but are not expected to develop the cancer?
(A) 1/400
(B) 19/400
(C) 20/400
(D) 38/400
Let's think step by step: The expected proportion of individuals who carry the b allele but are not expected to develop the cancer equals to the frequency of heterozygous allele in the given population. According to the Hardy-Weinberg equation p^2 + 2pq + q^2 = 1, where p is the frequency of dominant allele frequency, q is the frequency of recessive allele frequency, p^2 is the frequency of the homozygous dominant allele, q^2 is the frequency of the recessive allele, and 2pq is the frequency of the heterozygous allele. Given that q^2 = 1/400, hence, q = 0.05 and p = 1-q = 0.95. The frequency of the heterozygous allele is 2pq = 2*0.05*0.95 = 38/400.
The correct answer is (D)

Question: A Fe pellet of 0.056g is first dissolved in 10mL of hydrobromic acid HBr (0.1M). The resulting solution is then titrated by KMnO4 (0.02M). How many equivalence points are there?
(A) Two points, 25ml and 35ml
(B) One point, 25mL
(C) One point, 10ml
(D) Two points, 25ml and 30ml
Let's think step by step: HBr reacts with Fe to produce Fe2+. MnO4- initially reacts with Fe2+ followed by Br-. There are two equivalence points at 25ml and 35ml. In the beaker, the present species are Fe2+ and Br-. In a titration involving two analytes, it's essential to identify which reaction occurs first. Given the redox nature of the titration and the reduction potentials: E0 (Br2/Br-) = 1.09V, E0 (MnO4-/Mn2+) = 1.49V, and E0 (Fe3+/Fe2+) = 0.77V. With [Fe2+] determined as 0.1M, two reactions are considered. Reaction 1: MnO4- reacts with 5Fe2+ and 8H+ to produce Mn2+, 5Fe3+, and 4H2O. Reaction 2: 2MnO4- reacts with 10Br- and 16H+ to produce 2Mn2+ and 5Br2 with 8H2O as a byproduct. MnO4- first reacts with Fe2+ in a 1:5 ratio, making the first equivalence point at 25ml. Once Fe2+ is exhausted, MnO4- reacts with Br- in a 2:10 ratio, adding another 10ml for a total second equivalence point at 35ml.
The correct answer is (A)

Question: Consider a quantum mechanical system containing a particle of mass m moving in an isotropic three dimensional potential of the form V(r) = 1/2mω^2r^2 corresponding to the acted force obeying Hooke's law. Here, ω is the angular frequency of oscillation and r is the radial distance of the particle from the origin in spherical polar coordinate. What is the value of energy of the third excited state, and how many linearly independent eigenfunctions are possible for the same energy eigenvalue?
(A) 11 π^2ℏ^2/(2mr^2), 3
(B) (9/2) ℏω, 10
(C) 11 π^2ℏ^2/(2mr^2), 10
(D) (9/2) ℏω, 3
Let's think step by step: This problem is nothing but the three dimensional simple harmonic oscillator (SHO) problem. The energy spectrum of three dimensional SHO is En = (n+3/2)ℏω where n = 0,1,2,3..... For third excited state n=3. 3+3/2=6/2+3/2=9/2. Thus the corresponding energy is (9/2)ℏω. The degeneracy of the state is gn = (n+1)(n+2)/2. For n=3, degeneracy is (3+1)*(3+2)/2=4*5/2=10.
The correct answer is (B)

Question: Your overhear two chemists talking to each other as they leave a synthetic organic
chemistry lab. One asks the other "So, how did it go?" The second chemist replies, "Not well -
my compounds are on top of each other." What is the second chemist most likely referring to?
Choices:
(A) The compounds they are working with have similar polarities.
(B) The compounds they are working with have similar boiling points.
(C) The compounds they are working with are bonding to each other through non-covalent/van
der Waals interactions.
(D) The compounds they are working with have similar optical rotations.
Let's think step by step:
"On top of each other" commonly refers to two compounds that have similar Rf values on
chromatography (a common operation in synthetic chemistry). Similar Rf values arise for
compounds with similar polarities.
The correct answer is (A)

Question: Mitochondria are semi-autonomous cellular organelles in charge of energy production.
They encode for a part of their own translational machinery and respiratory complexes. Mito-
chondrial function is governed by over a thousand proteins imported from the cell, contributing
to processes like the transport of proteins, ribosome biogenesis and translation regulation,
respiratory oxidation, metabolism, and apoptotic signaling cascade. Mutations in the code
for mitochondrial protein networks can cause numerous diseases in humans that are inherited
through generations. Mutations of which of the mitochondrial proteins listed below are least
likely to be genetically transmitted from a father to his children?
Choices:
(A) Translocase of inner mitochondrial membrane 17B
(B) ATP binding cassette subfamily B member 8
(C) NADH dehydrogenase 2
(D) Tu translation elongation factor, mitochondrial
Let's think step by step: The colleague should know that mitochondria from fathers are rarely if
ever, transmitted to their offspring. Therefore, the protein encoded by the paternal mitochondrial
genome will most likely not be passed down the generation. NADH dehydrogenase 2 is the only
one encoded by the mitochondrial genome from the MT-ND2 gene among the listed proteins.
Leigh's syndrome, lactic acidosis, and metabolic diseases are all linked to a mutation in the ND2
gene. ATP binding cassette subfamily B member 8 (ABCB8) is a chromosome 7 encoded gene;
Tu translation elongation factor, mitochondrial is chromosome 16 gene TUFM. Translocase
of inner mitochondrial membrane 17B is chromosome X coded gene TIMM17B. There is no
evidence that it is maternally imprinted; hence, daughters may inherit the father's gene copy in a
50:50 ratio.
The correct answer is (C)

Two quantum states with energies E1 and E2 have a lifetime of 10^-9 sec and 10^-8 sec, respectively. We want to clearly distinguish these two energy levels. Which one of the following options could be their energy difference so that they can be clearly resolved?

Choices:
(A) 10^-8 eV
(B) 10^-11 eV
(C) 10^-9 eV
(D) 10^-4 eV

Give step by step reasoning before you answer, and when you're ready to answer, please use the format "The correct answer is (insert answer here)".

We use the same answer regex as the original paper to check for correctness:

answer is \((C)\)|Answer: \((C)\)|answer: \((C)\)|answer \((C)\)