Repository navigation
feat(models): condense memory and default helpers to Haiku 5.5 - #44
Merged
Merged
Conversation
added 3 commits
October 7, 2026 21:42
Haiku 5.5 (`claude-haiku-5-5`) is out at a tenth of Haiku 4.5's price ($0.10 in / $0.50 out / $0.01 cache read per million tokens, for prompts up to 100K tokens), but the claude model picker could not offer it, because the catalog stopped at Haiku 4.5. Added to the claude list in both the shipped catalog and the published remote copy, directly above Haiku 4.5, which stays selectable. The catalog test's exact list carries the new row; it was red until the entry existed.
The memory condenser and the default model for narrow helper agents (triage, routing, verification, formatting) were pinned to Haiku 4.5. Haiku 5.5 does the same job at a tenth of the price, and both are cheap, high-volume calls: the condenser summarizes evicted memory sections on a schedule, and its verify gate still rejects a summary that is not smaller or that drops the required shape. CONDENSE_MODEL in reflect.ts and MODEL_HELPER in config.ts now name `claude-haiku-5-5`. An explicit per-agent model choice still overrides the helper default, as before. Tests assert the model each path actually uses: the condenser's hidden session is started with claude-haiku-5-5, and every helper role resolves to it. Both check the id is in the shipped catalog, because an id the CLI does not know fails every call. Both were red against the 4.5 pins.
priceFor matched every model id containing "haiku" to one row
($0.80 in / $4 out per million tokens). When a session's own cost is
missing, a Haiku 5.5 session was therefore estimated at about eight
times its list price ($0.10 in / $0.50 out / $0.01 cache read / $0.125
cache write).
A HAIKU_5_5 row is matched before the general Haiku row; every other
Haiku id keeps the price it had.
Test: priceFor('claude-haiku-5-5') returns the 5.5 rates and a million
tokens of each kind costs $0.735; a Haiku 4.5 id is unchanged. Red
before the row existed.
🚫 This PR is missing its before/after evidenceEvery pull request here has to show its work. Screenshots or a short screen recording, before the change and after it.
How to fix it: edit the description, keep the A bug fix with no visible surface still needs it: show the failing behaviour, then the same steps passing. A terminal recording is fine. Genuinely nothing to show — a CI tweak, a typo, a dependency bump? A maintainer can apply the |
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
1 participant
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 the memory condenser and the default helper model from Haiku 4.5 to Claude Haiku 5.5 (
claude-haiku-5-5), a tenth of the price ($0.10 in / $0.50 out / $0.01 cache read per MTok).reflect.tsCONDENSE_MODELandconfig.tsMODEL_HELPERnow point atclaude-haiku-5-5; an explicit per-agent model still wins.src/shared/modelCatalog.jsonanddocs/model-catalog.json; Haiku 4.5 stays selectable.pricing.tsgains a Haiku 5.5 fallback row, matched before the general Haiku row, so estimates are not 8x high when per-model cost is missing.Takes effect after a rebuild and relaunch.