On the same day, OpenAI engineer Thibault Sottiaux (Tibo) announced that the default context length for GPT-6.1 Sol has been increased from 272k tokens to 500k tokens in Codex and Work environments, an increase of about 80%.

This change means that the amount of text the model can process in a single conversation is almost doubled—whether it's handling an entire codebase, reading long reports, or maintaining complex tasks across multiple rounds with detailed instructions, it's less likely to trigger automatic compression, and conversations are less likely to be abruptly cut off or forced into summaries. For example, tracing the source of clauses in a hundreds-page contract, or having AI continuously follow an evolving development project with constantly added requirements, a longer window means fewer "memory lapses" and less need for repeated explanations. For users who heavily rely on long-term memory, the experience improvement is very direct and reflects the industry competition around "window" size in large models.

Manually adjustable up to 1M, no extra charge for long texts this time

Tibo explained that 500k is the "sweet spot" the team found after repeated testing for most users in terms of experience and performance, but it is far from the limit: users can still manually adjust, and the maximum upper limit supported by GPT-6.1 Sol has now reached 1M.

The more critical change is on the billing side—the adjustment comes with OpenAI no longer charging separately for long context. In the past, such long windows were considered high-cost add-ons. Now, with the default increased and no additional fees, it effectively lowers the threshold for heavy usage. At a time when long context has become a must-have selling point for all players, directly passing the expansion benefits to users is also a practical competitive move by OpenAI.