Summary
Claudian 2.0.30 added dynamic Codex model discovery in f1516a7, which should let GPT-5.6 variants and their model-specific reasoning efforts flow from the Codex app-server into the picker.
One advertised effort is still deliberately removed:
const EXCLUDED_REASONING_EFFORTS = new Set(['ultra']);
in src/providers/codex/models.ts.
Observed Codex catalog
With Codex CLI 0.141.0, the app-server/model catalog available to my account reports:
| Model |
Advertised reasoning efforts |
Default |
gpt-5.6-sol |
low, medium, high, xhigh, max, ultra |
low |
gpt-5.6-terra |
low, medium, high, xhigh, max, ultra |
medium |
gpt-5.6-luna |
low, medium, high, xhigh, max |
medium |
The catalog describes ultra as maximum reasoning with automatic task delegation. Claudian currently filters it during normalization, so Sol and Terra cannot select it even though the provider advertises it.
Request
Please make Codex reasoning efforts capability-driven all the way through:
- expose
ultra for models that advertise it, currently Sol and Terra;
- do not expose it for models that do not advertise it, such as Luna;
- preserve the app-server descriptions and model-specific defaults;
- forward the selected effort through Codex thread/turn requests as with the existing levels.
If ultra is intentionally excluded because its automatic delegation semantics are incompatible with Claudian, please document that limitation and the specific incompatibility.
Suggested acceptance checks
- A discovered Sol or Terra model shows
Low, Medium, High, XHigh, Max, and Ultra.
- A discovered Luna model stops at
Max.
- Selecting
Ultra survives provider switching and conversation resume.
- Unit tests cover catalogs where only some models advertise
ultra.
Thanks for adding dynamic model discovery—this request is only for the remaining model-advertised effort level.
Summary
Claudian 2.0.30 added dynamic Codex model discovery in
f1516a7, which should let GPT-5.6 variants and their model-specific reasoning efforts flow from the Codex app-server into the picker.One advertised effort is still deliberately removed:
in
src/providers/codex/models.ts.Observed Codex catalog
With Codex CLI 0.141.0, the app-server/model catalog available to my account reports:
gpt-5.6-sollow,medium,high,xhigh,max,ultralowgpt-5.6-terralow,medium,high,xhigh,max,ultramediumgpt-5.6-lunalow,medium,high,xhigh,maxmediumThe catalog describes
ultraas maximum reasoning with automatic task delegation. Claudian currently filters it during normalization, so Sol and Terra cannot select it even though the provider advertises it.Request
Please make Codex reasoning efforts capability-driven all the way through:
ultrafor models that advertise it, currently Sol and Terra;If
ultrais intentionally excluded because its automatic delegation semantics are incompatible with Claudian, please document that limitation and the specific incompatibility.Suggested acceptance checks
Low,Medium,High,XHigh,Max, andUltra.Max.Ultrasurvives provider switching and conversation resume.ultra.Thanks for adding dynamic model discovery—this request is only for the remaining model-advertised effort level.