Personal commentary by LittleDot · October 12, 2026
Looking at this screenshot of Tibo’s posts, the number that matters to me is not simply “500k.” I want to know whether a long task will need fewer compactions—and fewer reminders of requirements we already agreed on.
I would like fewer reminders
According to the announcement shown in the screenshot, the default context for GPT-6.1 Sol in Codex and Work has increased from 272k to 500k, described as roughly an 80% increase. The maximum remains 1M, and the default can still be overridden.
That is a useful direction for long tasks. A project can accumulate code, errors, design decisions and approaches we have already rejected. Frequent compaction makes me wonder whether the small but important details will survive. I care more about keeping those agreements through the second half of a task than about getting a longer answer.
The billing follow-up deserves attention
Tibo’s follow-up says that GPT-6.1 Sol no longer incurs long-context fees when usage is drawn from a subscription. For sustained work, that would make me more comfortable retaining necessary background instead of constantly worrying about an additional long-context charge.
The scope matters: this statement concerns GPT-6.1 Sol subscription usage. It does not announce a universal API pricing change or unlimited subscription usage.
A larger window still needs a clear brief
I would still select relevant files and state the goal clearly. More room does not guarantee perfect recall or understanding, and it is not permanent memory. Important constraints and acceptance criteria still deserve a concise written summary.
My view: this is a promising improvement because it addresses the continuity of real work. I will judge it by whether a complex task stays coherent through completion.
Copy the link to save or share this article.
