Experiments
Stop when this page is not the decision
A published page can record a stop. It does not, by itself, record the next plan.
x.com published this account (2026-10-11): Default context length for GPT-6.1 Sol increased from 272k to 500k when used via Codex and ChatGPT Work; long-context fees not charged on subscription usage. Max remains 1M; override still available.
This guide uses that public page as the example. This site did not run the test, close the deal, or measure the product.
Write only what the page says
The page's claim is: Default context length for GPT-6.1 Sol increased from 272k to 500k when used via Codex and ChatGPT Work; long-context fees not charged on subscription usage. Max remains 1M; override still available.
Put the URL under that sentence. If a later recap adds a number or a date, keep the original page as the record.
A published page can record a stop. It does not, by itself, record the next plan.
Do the site's check on this one page. Fill a line only when the page supports it. Where the page is silent, write Not stated.
A finished note names the page, the date, and the limit of the claim. It does not borrow a result from a different story.
Stop at the untested line
Anything you did not open, install, or count stays untested. Leave that line visible.
Source page: OpenAI raises GPT-6.1 Sol default context to 500k in Codex and ChatGPT Work.
Sources: OpenAI raises GPT-6.1 Sol default context to 500k in Codex and ChatGPT Work