[Transient-Onlooker]

Summary

• Add an opt-in, capability-gated Flex service tier to Codex's routed model catalog, alongside existing Fast and Ultra Fast.
• Enable Flex/Fast for the exact NanoGPT openai/gpt-6-luna model based on NanoGPT's advertised service_tier support; leave unrelated NanoGPT models unclassified.
• Apply the same metadata to live /v1/models discovery and both persisted catalog writers. Withdraw generated Flex metadata when the declaration is removed.
• Keep the model's default tier unchanged. This PR does not implement a forced per-model default/override.

Policy and compatibility

• An exact model must be Fast-eligible and have a service-tier wire with canonicalToWire.flex = "flex", foreignCallerTiers = "verbatim", and actual caller-tier forwarding permission.
• For third-party openai-chat providers, chatServiceTier: true is needed to carry the caller-selected foreign Flex tier; model capabilities alone do not imply safe forwarding.
• Unsupported and unclassified providers, combos, native/account models, and conflicting duplicate declarations do not acquire Flex support.
• fastMode still follows the existing global contract. Keep it Auto so a caller-selected Flex tier can pass through. Forced Fast or Default continues to override it.
• Flex can be slower and may return resource-unavailable errors; provider acceptance and billing are not guaranteed by a picker entry.

Changes

• Reuse the captured FastWire authority in catalog capability hints and strict final-wire eligibility checks.
• Add an exact-slug managed Flex descriptor, model hint propagation, and deterministic catalog merge inputs.
• Add focused regression cases for eligible/unsupported models, duplicate scopes, NanoGPT's exact model, and direct catalog projection.
• Document the feature in English and Korean, with the structure invariant updated.

Verification

• Verified source references and existing FastWire/catalog design.
• Updated all related files in one commit, based on upstream dev 1025acd4b.
• Not run locally: Bun build, typecheck, test suite, or live NanoGPT request. This environment cannot clone the GitHub repository or execute Bun. Tests added are not claimed to pass.
• Hosted CI and maintainer review pending.
• Confirm live Codex picker behavior and wire service_tier: "flex" on a configured NanoGPT key.

Related: #1086, #1886. This is selectable Flex metadata, separate from the previously proposed per-model tier override policy.

References: https://nano-gpt.com/blog/shipped-in-may-2026 · https://github.com/openai/codex/blob/main/codex-rs/tui/src/chatwidget/service_tiers.rs

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

• Required local validation passed; commands, results, and any full-suite exception are documented.

• I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).

• I resolved all correct Codex and CodeRabbit findings.

• My PR is ready for review.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[github-actions]

⏳ DRAFT

• review readiness checklist open (0/4 boxes ticked).

What to do

• Tick all four boxes in the PR description once you're done (currently 0/4).

Review readiness checklist

• ⬜ Required local validation passed; commands, results, and any full-suite exception are documented.
• ⬜ I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).
• ⬜ I resolved all correct Codex and CodeRabbit findings.
• ⬜ My PR is ready for review.

0/4 boxes ticked.

This PR stays in draft until every box above is ticked.

[lidge-jun]

Closing per a maintainer product decision for the provider sweep: the Flex tier stays opt-in, so #7534 ("expose opt-in Flex tier for eligible routed models") is the one we keep and review. This PR advertises Flex by default, which that decision rules out. Thank you for both versions; the work continues in #7534.