Engineering productivitytactic

Enable Codex 1M context window experimentally via `config.toml`

Problem solved: Codex defaults to a 270k context window even on GPT-5.4, which forces splitting large codebases into microservice-shaped chunks; the 1M window is available but opt-in.

TierCCatalog — Pattern outlined, candidates for deeper work

EffortS · Days

Forengineer with large codebases on Codex / GPT-5.4

What

GPT-5.4 supports 1M context experimentally; activation is via model_context_window and model_auto_compact_token_limit in Codex's config.toml (both CLI and GUI app inherit the same config). One reliable path: paste a request to the Codex agent itself asking it to enable the extended context via settings, citing the official 1M-context announcement. Billed at 2× normal rate. Only newly-opened chats inherit the larger window; existing sessions stay at 270k.

← All casesC-093