CONFIRMED
OpenAI nearly doubled, in one move, the memory space GPT-6.1 Sol uses by default inside Codex and ChatGPT Work, the two environments built for people coding and running long work sessions every day. The jump went from 272,000 to 500,000 tokens, the text units the model processes each turn, an increase of roughly 80%. In practice that means fewer interruptions mid session, because the model now takes much longer to hit the point where it needs to summarize what already happened and risk losing nuance along the way.
Why that number matters so much in real use
Every language model has a limit on how much text it can see at once, called a context window. When a long conversation blows past that limit, tools like Codex, OpenAI's coding agent, run what is called compaction, an automatic summary that trades fine detail for free space. It works, but it has a cost: a specific instruction, a variable name, an architecture decision made twenty messages earlier can vanish in the summary. Raising the default from 272,000 to 500,000 tokens pushes that summarization moment much further out, which matters a lot for anyone reviewing an entire codebase, chasing a rare bug for hours, or processing a huge corporate document without breaking it into pieces.
Who signed off on this change and what it covers
The announcement came straight from the person running the product: Thibault Sottiaux, known as Tibo, the OpenAI executive who leads Codex and ChatGPT Work, posted on X on October 11 that GPT-6.1 Sol's default, the mid tier model in the GPT-6 family built for coding and professional tasks at a fifth of flagship Astra's price, rose from 272,000 to 500,000 tokens. He called the new figure the tested sweet spot for most users and said the ceiling stays at 1.05 million tokens for anyone who manually overrides the limit. In a second post, Tibo confirmed that usage drawn from a Plus, Pro, Business, Enterprise or Edu subscription inside Codex and Work no longer pays the long context fee that existed before.
What still looks unfinished about this announcement
As of this report, neither Codex's official changelog nor OpenAI's pricing page listed the change, according to independent trackers that follow the product hour by hour. The announcement lives only in Tibo's two personal X posts, even though he speaks for the company and the technical detail matches the model's official page. And a bigger context window is not automatically the same as a better answer: more room postpones the summary, but it does not guarantee the model will use that space precisely, nor does it remove the need for manual compaction when a task demands it.
The bill that changes and the one that does not
The fee waiver only applies to subscription usage inside Codex and Work. On the raw, per token API, the doubled price rule for input above 272,000 tokens still stands, and feeding the model 500,000 tokens at once can cost substantially more than before, close to three times as much in a simple calculation with 10,000 output tokens. So the real win from this announcement goes to people paying a flat monthly subscription and running long work sessions, not to those billed by the token processed.
Sources
Thibault Sottiaux (OpenAI): announcement of GPT-6.1 Sol's new default context; Thibault Sottiaux (OpenAI): end of the long context fee on subscription usage; OpenAI: official GPT-6.1 Sol model documentation; AI Industry Today: independent cost breakdown and coverage of the change.
