MeetNewAI / Utilities

GPT IQ index

Radar Bench scores a percentage times 1.5, with a maximum of 150. 64 tasks cover software engineering, spatial reasoning, computer vision, terminal tools, systems engineering, scientific computing, and data analysis.

Source codexradar.com · cached 10/16/2026, 10:00

Bar = score / 150$ API-equivalent costMinutes = average timeCorner = graded tasksOutlined cell = highest score in the row

GPT-6 Astra

ULTRA64 tasks

111.6

MAX86 tasks

111.3

XHIGH63 tasks

121.1

HIGH64 tasks

112.6

MEDIUM75 tasks

90.6

LOW55 tasks

97.2

GPT-6.1 Sol

ULTRA55 tasks

105.0

MAX70 tasks

102.7

XHIGH61 tasks

90.0

HIGH76 tasks

99.6

MEDIUM64 tasks

75.1

LOW86 tasks

81.8

GPT-6 Sol

ULTRA55 tasks

68.0

MAX55 tasks

67.0

XHIGH60 tasks

49.8

HIGH66 tasks

56.2

MEDIUM65 tasks

35.8

LOW67 tasks

32.8

GPT-5.6 Sol

ULTRA40 tasks

93.8Thin data · 36/64 tasks

MAX60 tasks

74.4

XHIGH40 tasks

75.6Thin data · 35/64 tasks

HIGH51 tasks

64.1Thin data · 44/64 tasks

MEDIUM50 tasks

41.5

LOW48 tasks

29.8Thin data · 44/64 tasks

GPT-6 Luna

—

MAX71 tasks

52.1

XHIGH68 tasks

47.3

HIGH74 tasks

36.7

MEDIUM69 tasks

25.7

LOW73 tasks

11.9

GPT-5.6 Luna

—

MAX55 tasks

48.1Thin data · 39/64 tasks

XHIGH53 tasks

52.5Thin data · 42/64 tasks

HIGH54 tasks

39.0Thin data · 41/64 tasks

MEDIUM45 tasks

28.4Thin data · 34/64 tasks

LOW45 tasks

13.0Thin data · 40/64 tasks

Cost is the mean official Standard API equivalent. Time is the mean of known samples. The score uses the latest three valid grades per task. Coverage under three quarters of the bank is marked thin.