fix(core): hide Codex models ChatGPT accounts reject - #58180

Issue for this PR

None

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Problem: With the ChatGPT browser or headless sign-in (Codex backend), we explicitly allowed gpt-5.5 and gpt-5.3-codex-spark. The Codex backend rejects both for ChatGPT accounts ("not supported when using Codex with a ChatGPT account"), so the model picker offered models that always fail. Spark also got the blanket 400k/272k Codex cap, which is larger than its real 128k/100k window.

Change: Remove the allowlist and raise the version cutoff from "newer than 5.4" to "newer than 5.5". That also covers gpt-5.5-pro, so it no longer needs an entry in the blocklist.

-const codexAllowed = new Set(["gpt-5.5", "gpt-5.3-codex-spark"])
-const codexDisallowed = new Set(["gpt-5.5-pro", "gpt-5.6"])
+const codexDisallowed = new Set(["gpt-5.6"])

-if (!codexAllowed.has(apiID) && (codexDisallowed.has(apiID) || !match || !(major > 5 || (major === 5 && minor > 4))))
+if (codexDisallowed.has(apiID) || !match || !(major > 5 || (major === 5 && minor > 5)))

Out of scope:

• The ChatGPT token-sharing sign-in (chatgpt.ts) still lists gpt-5.5 and gpt-5.5-fast as fallbacks. It uses the API endpoint rather than Codex, and we haven't seen it reject them.
• API-key connections still show the full catalog.

How did you verify your code works?

• bun test test/plugin/provider-openai.test.ts in packages/core: 5 pass. The ChatGPT filtering test now checks that gpt-5.5 and Spark are hidden, and checks cost, headers, and caps on gpt-5.6-sol.
• bun typecheck in packages/core: clean.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

Added by the opencode bot (AI-generated).

Related: #47646