OpenAI Codex lead Tibo Sottiaux announced that users who sign into Codex with a ChatGPT account can now manually expand GPT-5.6 Sol's context to 1 million tokens. The setting was already public, but previously only worked with an API key.
That million is the context budget Codex allocates to the model — not a sudden capacity boost for GPT-5.6 Sol itself. The model natively supports up to 1.05 million tokens; Codex's default previously used only a portion, starting at around 372,000 tokens before being reduced to 272,000.
The model's capability and how much Codex exposes by default are two different things. GPT-5.6 Sol's official spec still lists 1.05 million tokens of context.
Users who need extra-long context can now manually set Codex's window to 1 million tokens and put the auto-compaction threshold at 900,000. That lets Codex hold onto more code, tool output, and chat history — it won't start compacting older content until you're nearing 900,000 tokens.
Sottiaux still recommends most people stick with the defaults, saying the current settings are already tuned for performance and cost.
https://twitter.com/thsottiaux/status/2089082893804896524