Possible automatic Astra selection and runtime metadata mismatch (single observed case)

What version of Codex CLI is running?

0.145.0-alpha.2 (affected scheduled CLI); 0.153.0 model-cache client observed later

What subscription do you have?

ChatGPT subscription (plan unknown)

Which model were you using?

gpt-6-astra (unexpectedly selected); gpt-5.6-terra was the prior scheduled default

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What terminal emulator and version are you using (if applicable)?

No response

Codex doctor report

What issue are you seeing?

I observed a possible automatic model-selection / runtime-metadata mismatch on a single Windows machine and account. I do not know whether other users are affected.\n\nOn 2026-09-05 KST, a VS Code voice session and a scheduled Codex exec job started as gpt-6-astra / low despite no explicit -m flag or model-selection instruction in the scheduled prompt. Prior scheduled jobs used gpt-5.6-terra / medium.\n\nThe scheduled 0.145.0-alpha.2 CLI then warned that priority was unsupported and Astra metadata was not found; an affected run returned HTTP 400 saying Astra required a newer Codex version. Later that day, a 0.153.0 model cache listed Astra and priority as supported.\n\nPrivate feedback was uploaded through /feedback. Thread ID: 01a07318-8157-7082-a533-2c3cc31341dd.

What steps can reproduce the bug?

1. Use an environment where scheduled Codex exec jobs previously ran with gpt-5.6-terra / medium. 2. On 2026-09-05 KST, observe a VS Code voice session and a scheduled Codex exec job with no -m flag or model-selection instruction. 3. Observe gpt-6-astra / low selected. 4. With the scheduled 0.145.0-alpha.2 CLI, observe metadata/service-tier warnings and HTTP 400 requiring a newer Codex version. This is one observed case only; I cannot provide a deterministic cross-account reproduction.

What is the expected behavior?

A scheduled job should retain an explicit or pinned model, or receive only a model supported by its installed runtime. A rollout or default change should not select an unavailable model, emit missing-metadata warnings, or cause a 400. It should be clear whether this is scoped to one account or a broader rollout.

Additional information

No response

Begin with the scheduled Codex exec path and the model-cache behavior, comparing the 0.145.0-alpha.2 and 0.153.0 observations. Determine whether model selection can choose an unsupported runtime model; done means scheduled jobs use a pinned or supported model without metadata warnings or HTTP 400 failures.