v2.5.0

Adds the GPT-5.6 family — Sol, Terra, and Luna — plus the two reasoning tiers above xhigh.

npm i -g codex-multi-auth

Additive and opt-in. gpt-5 still resolves to gpt-5.5, and the default model, quota probe, routing, rotation, storage, and auth are all unchanged.

The models

Model	Default effort	Notes
gpt-5.6-sol	low	Bare gpt-5.6 resolves here
gpt-5.6-terra	medium
gpt-5.6-luna	medium	No ultra tier

Sol's unusually low default is intentional — it matches upstream.

Two new reasoning tiers sit above xhigh: max on all three, and ultra on Sol and Terra only. None of the 5.6 tiers accept none or minimal; those get bumped up to a supported effort rather than sent and rejected.

ultra never reaches the wire

ultra means Codex's client-side subagent delegation, not a real backend effort level. Upstream rewrites it to max before sending, and we do the same — ultra stays selectable in config and on model ids, and goes out as max. The wire type excludes ultra outright, so a future change can't reintroduce it without a type error.

Two routing bugs fixed on the way

Codex Max was nearly rerouted. gpt-5.1-codex-max is a model id that happens to end in -max, not Codex at max effort. Adding max to the effort-suffix pattern carelessly would have stripped the suffix and resolved every Codex Max request onto plain gpt-5.1-codex.

Unrecognised gpt-5.6-* ids silently ran GPT-5.5. The general GPT-5 resolver has no minor version 6, so anything it didn't match fell through to the stable variant. A dedicated 5.6 resolver now claims those ids.

Where the model facts came from

The upstream Codex catalog and client source, not the published API docs. The docs pages contradict each other on whether a max tier exists, and the context window they advertise is for the API surface rather than the ChatGPT Codex backend this wrapper actually talks to.

Notes

• No fallback chain. An account without GPT-5.6 access fails the request rather than quietly degrading to 5.5.
• Pricing per 1M tokens: Sol $5 in / $30 out, Terra $2.50 / $15, Luna $1 / $6. Cached input is 10% of input.
• 5.6 routes to the gpt-5.2 prompt family — upstream ships no 5.6 prompt file, and gpt-5.5 already reuses that family the same way.