Skip to main content
BenchLM

GPQA Diamond (GPQA-D)

We show this table for reference; we do not rank on it.

Data verified 38 confirmed releases in the last 30 daysFollow model changes

A display-only GPQA Diamond reference from provider comparison charts.

Benchmark score on GPQA-D — October 1, 2026

We compile the GPQA-D rows from provider self-reports and secondary reports. GPT-6 Astra leads the table at 96.0%, followed by Sakana Fugu-Ultra (95.5%) and Sakana Fugu (95.5%). We do not use these results to rank models overall.

1Proprietary

GPT-6 Astra

OpenAI

96.0%
Overall 88.69Context 1.05M
2Proprietary

Sakana Fugu-Ultra

Sakana AI

95.5%
Context 1M
3Proprietary

Sakana Fugu

Sakana AI

95.5%
Context 1M

71 modelsKnowledgeCurrentDisplay onlyUpdated October 1, 2026

Benchmark score table (71 models)

Score
1
GPT-6 AstraOpenAI · Closed
96.0%
2
Sakana Fugu-UltraSakana AI · Closed
95.5%
3
Sakana FuguSakana AI · Closed
95.5%
4
GPT-5.6 SolOpenAI · Closed
94.6%
5
Gemini 3.1 ProGoogle · Closed
94.3%
6
Claude Opus 4.7 (Adaptive)Anthropic · Closed
94.2%
7
Claude Opus 4.8Anthropic · Closed
93.6%
8
GPT-5.5OpenAI · Closed
93.6%
9
Kimi K3Moonshot AI · Closed
93.5%
10
Step 5 PreviewStepFun · Closed
93.5%
11
GPT-5.6 TerraOpenAI · Closed
92.9%
12
GPT-5.4OpenAI · Closed
92.8%
13
Ornith-1.5-397BOrnith AI · Open weight
92.8%
14
Gemini 3.5 FlashGoogle · Closed
92.7%
15
Qwen3.8 MaxAlibaba · Open weight
92.6%
16
Qwen3.7 MaxAlibaba · Closed
92.4%
17
Pareto 26.10 PreviewUnbiased · Closed
92.4%
18
GPT-5.6 LunaOpenAI · Closed
92.3%
19
Hy4 previewTencent · Open weight
92.3%
20
Qwen3.8-Flash-NextAlibaba · Open weight
91.7%
21
GLM-5.2Z.AI · Open weight
91.2%
22
Qwen3.8-Omni-FlashAlibaba · Closed
91.0%
23
DeepSeek V4.1 FlashDeepSeek · Open weight
90.9%
24
Kimi K2.6Moonshot AI · Open weight
90.5%
25
Qwen3.7 PlusAlibaba · Closed
90.3%
26
DeepSeek V4 Pro 0813DeepSeek · Open weight
90.1%
27
Interfaze BetaInterfaze · Closed
89.9%
28
Inkling-SmallThinking Machines Lab · Open weight
89.5%
29
Muse SparkMeta · Closed
89.5%
30
Claude Opus 4.6Anthropic · Closed
89.2%
31
Qwen3.8-27BAlibaba · Open weight
89.2%
32
Ornith-1.5-35B-A3BOrnith AI · Open weight
89.2%
33
DeepSeek V4 Pro (High)DeepSeek · Open weight
89.1%
34
Solar Pro 4Upstage · Closed
89.0%
35
Grok 4.20xAI · Closed
88.5%
36
DeepSeek V4 Flash 0731DeepSeek · Open weight
88.1%
37
InklingThinking Machines Lab · Open weight
87.9%
38
Kimi K2.5Moonshot AI · Open weight
87.6%
39
DeepSeek V4 Flash (High)DeepSeek · Open weight
87.4%
40
Hy3 PreviewTencent · Open weight
87.2%
41
MiniMax M2.7MiniMax · Open weight
87.0%
42
Nemotron 3 UltraNVIDIA · Open weight
87.0%
43
Ornith-1.5-9BOrnith AI · Open weight
86.4%
44
Solar Open 2Upstage · Open weight
86.3%
45
GLM-5.1Z.AI · Open weight
86.2%
46
GLM-5Z.AI · Open weight
86.0%
47
Ternary Bonsai 2 27BPrism ML · Open weight
85.8%
48
A.X K2SK Telecom · Open weight
85.6%
49
Ling 3.0 FlashInclusionAI · Open weight
85.0%
50
MAI-Thinking-1Microsoft · Closed
84.2%
51
Ling 3.0 Flash FP8InclusionAI · Open weight
84.0%
52
K-EXAONE 2.0LG AI Research · Open weight
82.2%
53
Mercury 2.5Inception · Closed
79.0%
54
Gemma 4 12BGoogle · Open weight
78.8%
55
Trinity-Large-ThinkingArcee AI · Open weight
76.3%
56
Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA · Open weight
75.6%
57
Mercury 2Inception · Closed
73.6%
58
DeepSeek V4 ProDeepSeek · Open weight
72.9%
59
Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight
72.2%
60
DeepSeek V4 FlashDeepSeek · Open weight
71.2%
61
ZAYA1-8BZyphra · Open weight
71.0%
62
MiniCPM5-2BOpenBMB · Open weight
70.2%
63
LongCat-Flash-Lite-SparseMeituan · Open weight
69.5%
64
Trinity-Large-PreviewArcee AI · Open weight
63.3%
65
Mellum2-12B-A2.5B-ThinkingJetBrains · Open weight
57.6%
66
ZAYA1-74B-PreviewZyphra · Open weight
57.3%
67
LLaDA2.2-miniInclusionAI · Open weight
44.4%
68
Soofi S 30B-A3BSoofi Project · Open weight
43.4%
69
Mellum2-12B-A2.5B-InstructJetBrains · Open weight
40.9%
70
MiniCPM5-1BOpenBMB · Open weight
26.3%
71
LFM2.5-230MLiquidAI · Open weight
25.4%

Among the reported GPQA-D rows, GPT-6 Astra is first at 96.0%. The third row is 0.5 points behind. The broader top-10 range is 2.5 points, so many of the published results sit in a relatively narrow band.

71 models have been evaluated on GPQA-D. The benchmark falls in the Knowledge category. GPQA-D is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About GPQA-D

Year

2026

Tasks

Graduate-level science questions

Format

Multiple choice questions

Difficulty

Graduate level

BenchLM stores GPQA-D separately from the standardized GPQA row when providers publish exact chart values that should not overwrite the core weighted benchmark.

Trinity-Large-Thinking: Scaling an Open Source Frontier Agent

Freshness and provenance

Version

GPQA-D 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does GPQA-D measure?

A display-only GPQA Diamond reference from provider comparison charts.

Which model scores highest on GPQA-D?

GPT-6 Astra by OpenAI currently leads with a score of 96.0% on GPQA-D.

How many models are evaluated on GPQA-D?

71 AI models have published results on GPQA-D in the BenchLM catalog.

Compare top models on GPQA-D

GPT-6 Astra vs Sakana Fugu-UltraSakana Fugu-Ultra vs Sakana FuguSakana Fugu vs GPT-5.6 SolGPT-5.6 Sol vs Gemini 3.1 Pro
Last updated: October 1, 2026 · BenchLM version GPQA-D 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.