Benchmarks / Knowledge

GPQA Diamond (Vals)

Graduate-level Google-Proof Q&A benchmark evaluating models on questions that require deep reasoning. Scores are the values published by VALS AI. Source benchmark version 1.

120models
10%Open Index weight
2026-09-01last updated
Vals AIpublisher

GPQA Diamond (Vals)

Higher is better · best effort per model

Chart settings
120 of 120 selected

full table below

#ModelEffortScoreCost / taskHarness
1Gemini 3.1 Prohigh95.5%$0.081–
2GPT-5.6 Solmax95.2%$0.110–
3Grok 4.6high94.7%$0.045–
4Gemini 3.8 Flashhigh94.4%$0.056–
5Gemini 3.7 Flashhigh93.9%$0.023–
6Qwen 3.8 Maxmax93.7%$0.077–
7Gemini 3.6 Flashhigh93.4%$0.032–
8Opus 5max93.4%$0.095–
9Fable 5.1max93.4%$0.205–
10Fable 5max93.2%$0.450–
11GPT-5.5xhigh93.2%$0.083–
12Kimi K3max92.9%$0.088–
13Grok 4.5high92.9%$0.044–
14Gemini 3.5 Flashhigh92.7%$0.064–
15MiniMax M3max92.7%$0.025–
16Opus 4.8max92.4%$0.339–
17DeepSeek V4 Pro 0813max92.4%$0.066–
18GPT-5.6 Lunamax91.7%$0.011–
19Gemini 3 Prohigh91.7%$0.090–
20GPT-5.2xhigh91.7%$0.149–
21GPT-5.4xhigh91.7%$0.209–
22Grok 4.3xhigh91.4%$0.025–
23Muse Spark 1.1xhigh91.2%$0.034–
24GPT-5.6 Terraxhigh90.9%$0.034–
25Qwen 3.7 Maxmax90.2%$0.061–
26Opus 4.7max90.2%$0.365–
27DeepSeek V4 Flash 0731high89.9%$0.013–
28Muse Spark–89.6%$0.0031–
29Opus 4.6max89.6%$0.295–
30DeepSeek V4max89.4%$0.126–
31Kimi K2.6max89.1%$0.080–
32Sonnet 5max88.9%$0.078–
33Qwen 3.8 27Bxhigh88.9%$0.047–
34Grok 4.20xhigh88.6%$0.034–
35Grok 4high88.1%$0.148–
36GLM 5.3max88.1%$0.056–
37Gemini 3 Flashhigh87.9%$0.044–
38Qwen 3.5 Plusnone87.4%$0.019–
39Qwen 3.6 Plusmax87.4%$0.023–
40Inklingmax87.1%$0.166–
41GPT-5.1high86.6%$0.075–
42MiniMax M2.7high86.6%$0.014–
43GLM 5.3 Flashmax86.4%$0.0016–
44Nemotron 3 Ultra 550B A55Bmax86.1%––
45Opus 4.5high85.9%$0.713–
46GPT-5high85.6%$0.073–
47GLM 5.2max85.6%$0.057–
48Sonnet 4.6max85.6%$0.409–
49Grok 4 Fastnone85.4%$0.0030–
50Ling 3.0 Flash–84.8%$0.0033–
51Qwen 3 Maxmax84.8%$0.074–
52GLM 5.1max84.5%$0.045–
53Grok 4.1 Fasthigh84.3%$0.0029–
54o3high84.1%$0.036–
55Kimi K2.5max84.1%$0.030–
56Gemini 3.5 Flash-Litehigh83.8%$0.012–
57Inkling Smallxhigh83.6%$0.020–
58GLM 5high83.3%$0.034–
59GPT-5.4 Minixhigh83.1%$0.013–
60Qwen 3.5 Flashmax82.8%$0.0034–
61MiMo V2.5 Proxhigh82.6%$0.0067–
62MiniMax M2.5high82.1%$0.018–
63Sonnet 4.5high81.6%$0.114–
64Gemini 2.5 Flashmedium81.6%$0.0029–
65MiMo V2.5xhigh81.6%$0.0025–
66Gemini 3.1 Flash-Litehigh81.1%$0.0023–
67Gemini 2.5 Prohigh80.8%$0.013–
68GPT-5 Minihigh80.3%$0.014–
69DeepSeek V3.2high80.3%$0.0035–
70GLM 4.7none80.0%$0.025–
71Grok 3 Minihigh79.3%$0.0039–
72gpt-oss-120bhigh78.5%$0.0062–
73MiniMax M2.1–78.5%$0.011–
74Kimi K2medium78.5%$0.023–
75GPT-5.4 Nanohigh77.5%$0.0016–
76Opus 4.1high76.3%$0.145–
77o3 minihigh75.5%$0.036–
78Sonnet 3.7medium75.3%$0.189–
79Sonnet 4high75.0%$0.117–
80o4 minihigh74.5%$0.013–
81GLM 4.6none74.5%$0.015–
82Grok 3none74.2%$0.039–
83o1high73.2%$0.253–
84Haiku 4.5xhigh72.2%$0.055–
85GLM 4.5xhigh72.2%$0.019–
86Opus 4medium71.7%$0.067–
87Gemini 2.5 Flash Litenone70.2%$0.0005–
88Qwen 3 235B–70.2%$0.0065–
89Llama 4 Maverickmedium69.4%$0.0010–
90gpt-oss-20bhigh68.9%$0.0067–
91Mistral Large 3none68.4%$0.0022–
92GPT-4.1 Minihigh67.9%$0.0023–
93GPT-4.1high65.4%$0.014–
94Gemini 2.0 Flashmedium65.2%$0.0005–
95GPT-5 Nanohigh63.4%$0.0046–
96Llama 3.3none62.4%––
97Magistral Medium 1.2medium62.4%$0.036–
98DeepSeek V3none61.6%$0.0016–
99Claude 3.5 Sonnet Latest–59.3%$0.0091–
100MiMo V2 Flashhigh59.3%$0.0012–
101Gemini 1.5 Pronone58.3%$0.0037–
102Magistral Small 1.2–58.3%$0.0071–
103Laguna XS.2–55.0%$0.0014–
104GPT-4ohigh53.8%$0.0095–
105GPT-4.1 Nanohigh50.8%$0.0005–
106Mistral Small 02–50.8%$0.0006–
107Grok 2none50.8%$0.0095–
108Command Anone48.5%$0.010–
109Mistral Large 11–47.7%$0.0067–
110Llama 4 Scoutmedium47.0%$0.0007–
111Gemini 1.5 Flashnone46.0%$0.0002–
112Mistral Small 3.1none44.2%$0.0003–
113GPT-4o minihigh44.2%$0.0005–
114Haiku 3.5medium37.9%$0.0022–
115Jamba 1.6 Largenone36.1%$0.0067–
116Mistral Medium 3.5high34.8%$0.095–
117Jamba 1.6 Mininone32.1%$0.0005–
118Command-R+ (Apr '24)none31.1%$0.0074–
119GPT-3.5high30.6%$0.0009–
120Laguna M.1–27.0%––