cubic
Try cubic for free
Updated

New AI coding model launches: adoption and first bug rates

As of Nov 9, 2026, the newest AI coding model on Bug Bench is GPT-6.1 Sol, released Sep 29, 2026. Of the 7 launched in the last 90 days, GPT-6 Luna has the lowest bug rate (0.31, #1 overall).

Launched in the last 90 days

Line = weekly share of AI-written code

  • GPT-6.1 Sol

    Released Sep 29, 2026First seen week of Sep 28vs GPT-5.6 Sol →

    #2 · 0.74

    9.9% last week

  • Claude Sonnet 5.5

    Released Sep 28, 2026First seen week of Sep 28

    #5 · 1.04

    3.7% last week

  • Claude Opus 5.5

    Released Sep 22, 2026First seen week of Sep 21vs Claude Opus 5 →

    #6 · 1.11

    66.2% last week

  • GPT-6 Luna

    Released Sep 22, 2026First seen week of Sep 21

    #1 · 0.31

    0.7% last week

  • Grok 4.7

    Released Sep 21, 2026First seen week of Sep 21

    #7 · 1.61

    0.6% last week

  • GPT-6 Astra

    Released Sep 4, 2026First seen week of Aug 31

    #4 · 0.95

    4.5% last week

  • Claude Fable 5.1

    Released Sep 1, 2026First seen week of Aug 31

    #3 · 0.88

    2.8% last week

Questions and answers

What is the newest AI coding model on Bug Bench?
GPT-6.1 Sol, released Sep 29, 2026. It wrote 9.9% of the AI-written code reviewed by cubic last week.
Which AI coding model released in the last 90 days has the lowest bug rate?
GPT-6 Luna, at 0.31 (#1 overall).
Which new AI coding model is used the most?
Claude Opus 5.5, with 66.2% of the AI-written code reviewed by cubic in the week of Nov 2, 2026.