← All models

GPT-6 Luna editing test

OpenAI · Tested Oct 27, 2026 · 1m 20s · $0.02

59/90 caught

  • Grammar & spelling35/45
  • Repetitions11/15
  • Redundancy5/15
  • Consistency8/15

67 comments · 3 extra flags

GPT-6 Luna caught 59 of 90 planted mistakes, 14th of 16 models tested. It was strongest on grammar and spelling (35 of 45) and missed most on redundancies (5 of 15). The 4 checks took 1m 20s and cost $0.02; asked for all of them in one prompt, it caught 60.

Every comment GPT-6 Luna left on our 6,000-word test story, where it left it. Click a missed mistake to find it in the text. The checks come from icanwrite, a markdown editor for writers.

Missed · 31