Codex App on Bedrock hard-codes gpt-5.4-mini for background tasks, causing 404

What version of the Codex App are you using (From “About Codex” dialog)?

26.611.62324

What subscription do you have?

bedrock api billing

What platform is your computer?

macOS

What issue are you seeing?

PLEASE NOTE ID may be useless as I turned off the telemetry.

When Codex App is configured to use Bedrock, the main visible session can use openai.gpt-5.5, but Codex starts a background/non-visible task with gpt-5.4-mini.

That background request is sent to the Bedrock Responses endpoint and fails with 404 because the model is not available there.

Main visible session metadata:

• provider: amazon-bedrock
• model: openai.gpt-5.5
• reasoning effort: xhigh
• visible thread/session id: 019ed815-7441-7741-b541-003d71ca1558

Failing background session:

• session id: 019ed815-ab2d-7d10-ab45-fa69399749bb
• turn id: 019ed815-abff-7490-83e9-e668bca94745
• app version from logs: 0.140.0-alpha.19
• selected model in background task: gpt-5.4-mini
• request id: req_amoy4pllkxlgcdat5h6os6gkehftktcvznki7jtmlmd6dojmi3iq

Error from ~/.codex/logs_2.sqlite:

2026-06-17 17:16:12
Turn error: unexpected status 404 Not Found: The model 'gpt-5.4-mini' does not exist,
url: https://bedrock-mantle.us-east-1.api.aws/openai/v1/responses,
request id: req_amoy4pllkxlgcdat5h6os6gkehftktcvznki7jtmlmd6dojmi3iq

### What steps can reproduce the bug?


1. Configure Codex App to use Bedrock / Bedrock API billing.
2. Start a new Codex App session in a workspace.
3. Use a supported visible model, for example `openai.gpt-5.5`.
4. Send a normal prompt.
5. Inspect `~/.codex/logs_2.sqlite`.

Reproduction/evidence query:

```bash
sqlite3 -readonly ~/.codex/logs_2.sqlite "
select datetime(ts,'unixepoch','localtime') as time,
       level,
       target,
       replace(coalesce(feedback_log_body,''), char(10), ' ') as log
from logs
where feedback_log_body like '%gpt-5.4-mini%'
   or feedback_log_body like '%The model ''gpt-5.4-mini'' does not exist%'
order by ts desc, ts_nanos desc
limit 20;
"

What is the expected behavior?

When Codex App is configured for Bedrock, all foreground and background model calls should use a model available through the configured Bedrock provider.

If a background task needs a smaller summarization/title model, Codex should either:

• map it to a Bedrock-supported model, or
• use the same configured model/provider as the visible session, or
• skip the background task with a non-fatal warning instead of surfacing repeated 404 failures.

Additional information

This appears reproducible across more than one background session.

Second example:

2026-06-17 17:13:01
session id: 019ed812-bfdb-7ca3-97ad-6a71f5e3a7c1
turn id: 019ed812-c0bf-78a2-915b-d0a95a34cb22
model: gpt-5.4-mini

Turn error: unexpected status 404 Not Found: The model 'gpt-5.4-mini' does not exist,
url: https://bedrock-mantle.us-east-1.api.aws/openai/v1/responses,
request id: req_kbbpd4775m3zdiyl62cegf25ip4kmseg5hhowgbtchgagjwnucsa