Codex¶
The Codex harness runs OpenAI’s
Codex CLI inside the FLARE Docker
container, billing usage against your ChatGPT subscription via OAuth
credentials stored in ~/.codex.
Choosing a plan¶
Codex is available on paid ChatGPT plans. Compare plans at ChatGPT pricing and pick one that covers your expected usage, then monitor your consumption at Analytics.
Authenticating for FLARE¶
Authenticate the Codex CLI on the host following these instructions:
codex login
This writes credentials to ~/.codex. FLARE bind-mounts that
directory read-write into the Docker container at run time so Codex
can refresh its access token mid-session.
Using the harness¶
from milp_flare.harness import CodexHarness
harness = CodexHarness(model="gpt-5.4", effort="high")
Or via the HARNESSES registry (convenient for config-driven
experiment scripts):
from milp_flare import HARNESSES
harness = HARNESSES["codex"](model="gpt-5.4", effort="high")
Cost tracking¶
The Codex CLI does not report per-run USD, so the harness computes it from token totals using the pricing table in COST_PER_MTOK. This is used to populate cost_usd in HarnessRunResult. Ensure the pricing table reflects current OpenAI API prices.