Thibault Sottiaux from OpenAI released a change on October 12 that will relieve heavy code users: in Codex and Work environments, the default context window of GPT-6.1Sol has been increased from 272,000 tokens to 500,000 tokens, effectively expanding the model's "short-term memory" by about 80%.
Thibault explained that although OpenAI will continue to increase this limit in the future, this expansion is sufficient to significantly reduce the frequency of triggering context compression - the awkward moment when the model, unable to hold all the information, hastily summarizes and discards details. With a wider window, tasks that require holding a large amount of context for a long time and are full of complex details can be handled more easily by the model without frequently having to manually clean up.
He emphasized that the default value of 500,000 tokens has been thoroughly tested and is at the optimal balance point between experience and performance for most users, the so-called "sweet spot" that has been talked about for a long time. Of course, if someone feels it is not enough or thinks it is too wasteful, the manual knob is still open: GPT-6.1Sol supports a maximum context length of up to 1,000,000 tokens, and whether to set it to the maximum or reduce it is entirely up to the user.
