David Rein et al.

GPQA Diamond · BenchLM

The hardest subset of GPQA featuring the most challenging graduate-level science questions. Sometimes reported separately from the standard GPQA benchmark.

In index
Knowledge
Source ↗

Models65

Top result96.0%GPT-6 Astra

Top-3 spread0.5 pts

Year2023

Result and price

Results

65 results

#ModelResult

GPT-6 Astra96.0%

02SASakana Fugu95.5%

02SASakana Fugu-Ultra95.5%

GPT-5.6 Sol94.6%

Gemini 3.1 Pro94.3%

Claude Opus 4.7 (Adaptive)94.2%

GPT-5.593.6%

Claude Opus 4.893.6%

09STStep 5 Preview93.5%

09MAKimi K393.5%

GPT-5.6 Terra92.9%

GPT-5.492.8%

12OAOrnith-1.5-397B92.8%

Gemini 3.5 Flash92.7%

Qwen3.8 Max92.6%

Qwen3.7 Max92.4%

17TEHy4 preview92.3%

GPT-5.6 Luna92.3%

Qwen3.8-Flash-Next91.7%

GLM-5.291.2%

Qwen3.8-Omni-Flash91.0%

DeepSeek V4.1 Flash90.9%

23MAKimi K2.690.5%

Qwen3.7 Plus90.3%

DeepSeek V4 Pro 081390.1%

26INInterfaze Beta89.9%

Muse Spark89.5%

27TMInkling-Small89.5%

Qwen3.8-27B89.2%

Claude Opus 4.689.2%

29OAOrnith-1.5-35B-A3B89.2%

32UPSolar Pro 489.0%

Grok 4.2088.5%

DeepSeek V4 Flash 073188.1%

35TMInkling87.9%

36MAKimi K2.587.6%

37TEHy3 Preview87.2%

Nemotron 3 Ultra87.0%

MiniMax M2.787.0%

40OAOrnith-1.5-9B86.4%

41UPSolar Open 286.3%

GLM-5.186.2%

GLM-586.0%

44PMTernary Bonsai 2 27B85.8%

45STA.X K285.6%

46INLing 3.0 Flash85.0%

MAI-Thinking-184.2%

48INLing 3.0 Flash FP884.0%

49LAK-EXAONE 2.082.2%

50INMercury 2.579.0%

Gemma 4 12B78.8%

52AATrinity-Large-Thinking76.3%

Nemotron 3.5 Lightning 30B A3B NVFP475.6%

Nemotron 3 Nano Omni 30B A3B72.2%

55ZYZAYA1-8B71.0%

56OPMiniCPM5-2B70.2%

57MELongCat-Flash-Lite-Sparse69.5%

58AATrinity-Large-Preview63.3%

59JEMellum2-12B-A2.5B-Thinking57.6%

60ZYZAYA1-74B-Preview57.3%

61INLLaDA2.2-mini44.4%

62SPSoofi S 30B-A3B43.4%

63JEMellum2-12B-A2.5B-Instruct40.9%

64OPMiniCPM5-1B26.3%

65LILFM2.5-230M25.4%