[GPT-5.6 Sol Million-Token Context Now Open to ChatGPT Account Users, Previously Limited to API Key Users] According to monitoring by Beating, OpenAI Codex lead Tibo Sottiaux announced that users logging into Codex via their ChatGPT accounts can now manually set GPT-5.6 Sol's context window to 1 million tokens. Previously, this configuration was public but only effective for API Key users. The '1 million' here refers to the context budget allocated by Codex for the model, not an expansion of GPT-5.6 Sol's capacity. The GPT-5.6 Sol model itself supports up to 1.05 million tokens; however, Codex previously defaulted to using only a portion of this, initially around 372,000 tokens, later reduced to 272,000 tokens. The model's capability and Codex's default settings are two separate matters. GPT-5.6 Sol is still officially marked as supporting 1.05 million tokens of context. Users requiring ultra-long context can now manually set Codex's window to 1 million tokens and adjust the auto-compression threshold to 900,000 tokens. This allows Codex to retain more code, tool outputs, and chat history, only starting to compress older content when nearing 900,000 tokens. Tibo still recommends that most users stick to the default settings, stating that the current defaults have been optimized for performance and cost. [Original Link]