GPT-6.1 Sol's default context just got longer. On October 12, 2026, OpenAI's Tibo announced on X that the default context length for GPT-6.1 Sol in Codex and ChatGPT Work has been raised from 272k to 500k tokens, an increase of roughly 80 percent. Users can still override the setting, and the model's maximum context stays at 1M. Minutes later he added a second change: when GPT-6.1 Sol usage is drawn from a subscription allowance, the long-context surcharge no longer applies. One change is about capacity, the other about billing, and both are aimed at people running long tasks.
Why the default matters more than the maximum
The 1M maximum context was already on the spec sheet when GPT-6.1 Sol launched at the end of September, but what most people actually get is the product default. At the old 272k default, large codebases, long documents and multi-file jobs hit context compaction sooner: the system summarizes earlier content and continues, keeping the conclusions but often dropping details such as variable names and constraints. Raising the default to 500k means the model can hold far more code and material in a single task without any settings changes, and long sessions need far fewer lossy compressions along the way. Our earlier coverage of the GPT-6.1 Sol launch explains where the model sits and what it costs.
Long tasks such as coding agents benefit first
Coding agents that read a repository end to end, run tests and edit across many files are the clearest winners: when the context fits a whole module, the agent does not have to keep re-retrieving what it already read, and long chains of work stay coherent. Long-document review and multi-round data analysis gain the same way. Short questions and single-file edits will feel no difference. A word of caution, though: a bigger context is not an invitation to fill it. Stuffing in irrelevant files dilutes the model's attention on what matters and can slow responses down. Treat 500k as available capacity, not a target, and keep trimming tasks the way you did before.
No more long-context fee on subscriptions makes costs predictable
The second change is easy to scroll past, but heavy users may feel it more. Drawing long-context usage from a subscription allowance used to carry an extra charge, so the longer you ran, the more carefully you had to count. Removing it means subscription users of Codex and Work mostly pay through their plan allowance again, with no separate toll for crossing a context threshold. As rivals keep marketing million-token contexts, OpenAI is effectively moving long context from a paid extra toward a default capability.
For now the changes cover the defaults and subscription billing in Codex and ChatGPT Work. OpenAI has not said whether API defaults and pricing will follow, or whether other models get the same treatment. If you already run long tasks on GPT-6.1 Sol, the practical step is simple: update to the latest client, confirm the new default is active, and retry the jobs you previously had to split into pieces to dodge compaction.
