Independent newsroom

Friday, October 30, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

Labs · Friday, October 30, 2026

OpenAI says GPT-6 Luna now gives shorter, less agreeable answers

The File desk · Oct. 30, 2026, 6:37 a.m. ET

Partially confirmed1 primary sources

Status

Partially confirmed. OpenAI posted the October 29, 2026 addendum to the GPT-6 Luna system card. The behavior changes and the capability ratings are the company's account. This desk did not rerun the tests.
  • Confirmed
    The addendum was published October 29, 2026, and says the update gives more concise responses and reduces overly agreeable language.
  • Partially confirmed
    OpenAI says it is treating the update as High in cybersecurity and High for biological and chemical capability, with the same safeguards as the October 7 version.
OpenAI on Thursday posted an update to GPT-6 Luna, the model it says it put in ChatGPT on October 7. The company says the new version gives shorter answers and is less agreeable, and it calls the update GPT-6 Luna (October 29).1
The addendum says the update uses the same types of data and training as the October 7 version, and the same safeguards. Under the company's , OpenAI says it is treating the update as High in cybersecurity and High for biological and chemical capability. The model does not reach the company's High threshold in what it calls AI self-improvement. Those labels are OpenAI's.1
Why it matters (Estimate). Estimate. Someone using ChatGPT may get shorter and less agreeable answers from this version, if the description matches what shipped. The High labels are the company's names for risk areas, not a government certification. Higher factual-error rates, on OpenAI's own comparison, mean this update did worse than the October 7 version on those particular checks.
The addendum is full of the company's test tables, and this story does not reprint them. On one set of factuality checks, OpenAI says the updated model shows higher factual-error rates than the October 7 version on all six reported metrics, and still lower error rates than GPT-5.6 Luna on every metric it reports there.1
How we know

Notes

  1. 1.OpenAI Deployment Safety Hub, “Addendum to GPT-6 Luna System Card: October 29, 2026 Update,” published October 29, 2026. https://deploymentsafety.openai.com/gpt-6-luna-october-28-2026-addendum “Today, we’re introducing an update to GPT-6 Luna, which we released in ChatGPT on October 7. The updated model provides more concise responses and reduces overly agreeable language. We refer to the updated model as GPT-6 Luna (October 29).” Same types of data and training as GPT-6 Luna (October 7), and the same safeguards stack. “Under our , we are treating GPT-6 Luna (October 29) as High in cybersecurity and High for Biological and Chemical capability.” “The model does not reach our High threshold in AI Self-Improvement.” “On these factuality evaluations, the updated GPT-6 Luna shows higher factual-error rates than the original October version across all six reported metrics, while still maintaining lower error rates than GPT-5.6 Luna on every metric.” The biological evaluation tables on the page are the company's. They are not reprinted in the story.

What is still unknown or disputed

  • Whether every ChatGPT user has the update. The addendum says the model was released in ChatGPT on October 7 and describes an October 29 update. It does not print a rollout clock.
  • The six factuality metrics by name and score. The page discusses a figure. This desk is not copying the chart into the story.
  • An outside lab's confirmation of the High ratings. None was opened.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    Addendum to GPT-6 Luna System Card: October 29, 2026 Update

    OpenAI · October 29, 2026

    OpenAI says the October 29 update gives more concise responses and reduces overly agreeable language, and that it is treating the model as High in cybersecurity and biological and chemical capability.

    https://deploymentsafety.openai.com/gpt-6-luna-october-28-2026-addendum