Codex's 1M Context Window Is a Billing Trap, Not a Free Upgrade

A 1M context window sounds like a productivity unlock, but the 272k billing cliff means every large request costs at least double. Developers who turn this on without understanding the pricing model, or who run it through a proxy that lacks overage billing, will see sudden, steep charges.

Setting the 1M context window in Codex requires either a one-time CLI flag or a permanent entry in ~/.codex/config.toml. Both methods are straightforward, but the OpenCodex proxy tool overrides them. OpenCodex hardcodes a 372k context window in its source and injects its own config on every launch via a shim, so user edits to the TOML file disappear after each session. Stopping the proxy process is not enough; the auto-start mechanism must be disabled with `ocx system settings --auto-start off` before the native 1M setting sticks.

Even after the local config works, the server-side billing model undercuts the benefit. OpenAI treats any request exceeding 272k tokens as a premium tier: the entire input price doubles and the output price rises 1.5x. The 272k cap was originally a deliberate guardrail to prevent surprise charges, and the new 1M availability does not remove that pricing cliff. Subscribers now have access to the larger window, but overage billing remains separate and is not advertised alongside the feature announcement.

API proxy operators face the sharpest risk. If a proxy does not explicitly enable the separate billing switch for over-272k usage, a single user's 1M-context session can generate costs that the proxy owner absorbs directly.