feat(ai): use Luna as the default conversation model - #784
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository UI Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
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. Comment |
|
The latest updates on your projects. Learn more about Unkey Deploy
|
1 participant
Switch the Balanced conversation model from
openai/gpt-5.6-terratoopenai/gpt-5.6-luna. Dashboard Balanced requests, Slack answers, and shared agent calls without an override now select Luna. Public MCP remains a tools server whose external client chooses its own model. Quick/Deep selections and the separately configured Insights models are unchanged.Add Luna's standard credit-pricing rates from the AI Gateway catalog, verified September 11, 2026: $0.20 input, $1.20 output, $0.02 cache read, and $0.25 cache write per million tokens. Keep Terra pricing for remaining consumers. This retains the existing flat-rate accounting design; long-context and service-tier surcharges are not modeled by that design.
Validation: root lint passes, all 33 workspace typecheck tasks pass, and all 10 targeted model-default, usage-telemetry, and shared credit tests pass. The pre-push hook also passed the full 27-task workspace test run (AI package: 916 passed, 33 integration/environment skips, zero failures). The existing configured-model pricing test failed before adding Luna's rate entry; a new regression verifies fresh/cached token attribution and final credit math below the long-context threshold. The docs' normal generated collections were created before the successful full typecheck.
Rollout context: the preceding 64-conversation synthetic evaluation favored Luna (18/22 grounded numeric answers versus Terra's 16/22; 87.1% lower returned cost across 19 fully billed matched data pairs). That evaluation included the harness changes in #782 and #783, which are still open. Those measurements are not a benchmark of current staging; coordinate rollout with those fixes. This configuration slice is based directly on current staging, has no source dependency or changed-file overlap with those PRs, and does not include their harness changes. No production deployment performed.
AI-assisted implementation and verification with Codex, directed by a maintainer.
Summary by cubic
Switches the Balanced conversation model from
openai/gpt-5.6-terratoopenai/gpt-5.6-luna. Dashboard Balanced requests, Slack answers, and shared agent calls without an override now use Luna. Quick/Deep and Insights models are unchanged, and public MCP remains a tools server where the external client chooses its own model.Adds Luna's standard credit-pricing rates and a regression test covering fresh/cached token attribution and final credit math. The existing flat-rate accounting design is unchanged; long-context and service-tier surcharges are not modeled. Validation passed: lint, typecheck, and targeted tests. No production deployment; rollout should coordinate with the open harness changes from the preceding evaluation.
Written for commit edf4924. Summary will update on new commits.