Codex adds one-million-token context support for ChatGPT accounts
Codex now supports a one-million-token context window for ChatGPT account usage, according to Tibo’s post on X. The option previously worked only with API keys.
The configuration targets GPT-5.6 Sol, which has a documented context window of 1,050,000 tokens. A larger window allows Codex to retain more code, tool output and conversation history before automatically summarizing older material.
The recommended settings for ~/.codex/config.toml are:
model = "gpt-5.6-sol"
model_context_window = 1000000
model_auto_compact_token_limit = 900000
The settings belong at the top level of the file, before any section headers. After saving the file, Codex users are instructed to restart the client and begin a new session.
For a single command-line session, the same configuration can be supplied without changing the default file:
codex -m gpt-5.6-sol \
-c model_context_window=1000000 \
-c model_auto_compact_token_limit=900000
The model_auto_compact_token_limit value starts automatic history compaction at approximately 900,000 tokens, leaving some capacity below the one-million-token setting. A Codex session reportedly accepted the configuration in config.toml, while a configuration editor listed the same model and limits alongside high reasoning effort.
Tibo also cautioned that the existing Codex context length had been tuned for performance and cost, describing it as tuned to “~perfection.” A related Codex response reportedly warned that larger contexts can consume usage substantially faster.
Codex therefore exposes the larger window as an opt-in configuration rather than replacing its existing default. The supplied guidance recommends treating the one-million-token setting as a specialized option, particularly because retaining more material can increase usage.
Source: Tibo on X
