ci(warden): Use Luna 6 for auxiliary tasks - #217
Merged
Merged
Conversation
Switch the shared auxiliary model after a paired smoke comparison matched Luna 5.6 on dedup decisions and auxiliary tool use at lower reported cost. Preserve high reasoning effort. Co-Authored-By: GPT-6 Astra <noreply@anthropic.com>
gricha
marked this pull request as ready for review
September 22, 2026 19:39
dcramer
approved these changes
Sep 22, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Switch Warden’s shared auxiliary model from
openrouter/openai/gpt-5.6-lunatoopenrouter/openai/gpt-6-luna, preserving high reasoning effort. This covers deduplication and the other auxiliary helpers.A small paired smoke test against deployed Warden v0.49.0 matched on all 10 dedup checks per model (five labeled cases, two runs each), with no invalid responses or fallbacks. Both models also passed an auxiliary tool-use check. Luna 6 reduced median dedup time from 3.45s to 2.48s and reported cost from $0.0025768 to $0.0011098 across those 10 calls, about 57% lower. Fixtures were paraphrased from two existing regression cases with distinct-issue and location controls; this was not a production trace replay or a full auxiliary evaluation.
Both models correctly grouped batch paraphrases; Warden’s existing claim-preservation rule retained differently worded findings after grouping. Identical claims were merged. A fresh Pi model catalog resolves Luna 6; a locally cached catalog initially did not. The organization workflow uses fresh runners.