GPT-5.6 Sol百万上下文向ChatGPT账号用户开放，此前仅API Key能用

According to Descartes Beating monitoring, Tibo Sottiaux, the head of OpenAI Codex, announced that users logging into Codex through the ChatGPT account can now manually extend GPT-5.6 Sol's context to 1 million Tokens. Although this configuration was previously available, it only applied to API Key users.

In this context, the 1 million refers to the context budget used by Codex for the model, not a sudden expansion of GPT-5.6 Sol. The GPT-5.6 Sol model itself supports a maximum of 1.05 million Tokens; Codex previously only utilized a portion of this, initially around 372,000, which was later reduced to 272,000.

The model's capability and how much Codex opens by default are two different matters. The GPT-5.6 Sol official documentation still notes a 1.05 million Token context.

Users needing an extended context can manually set Codex's window to 1 million and set the auto-compression threshold to 900,000. This way, Codex can retain more code, tool outputs, and chat history, starting to compress old content only after reaching around 900,000 Tokens.

Tibo still recommends that the majority of users continue using the default settings. He mentioned that the existing defaults have been optimized for performance and cost.