Repository navigation
opencode: pool effort remap — wired rungs, capability order, meta refresh - #1010
Merged
Merged
Conversation
…ers, meta refresh From the 2026-10-26 pool effort audit (serve-resolved ladders + AA v4.3.2): - scout: all three routes #medium/bare -> #low. #medium silently pruned every ClinePass leaf (no medium on that ladder; GLM has no medium anywhere) and haiku-plan was unselected. #low is wired on every leaf of every door. - workhorse: meta/muse-spark-1.3 -> #low (was bare). Local probe: low is the highest Spark rung that separates from the default plateau (~40% less reasoning spend, quality-neutral on workhorse-class work); high ~= default. - specialist: reorder sonnet#high (47) -> sol#medium (48) -> pro#high (<=36) -> luna#max (38). Luna was second, a full capability band below the tier, ahead of band-correct Sol. - summit: sol #high -> #xhigh (51); muse-wonder-1.0 -> #max (52, was bare). - claude-code/claude-haiku-5-5: give the name-only variants bodies (thinking budgets 4096/8192/32768/65536 around the existing high=16384). Only high carried settings; low/medium/xhigh/max were labels that sent default requests. - meta provider: add the chat models the API actually lists (spark-1.2, spark-1.2-contributor, spark-1.3-contributor, wonder-1.0-contributor). Non-chat API models (sam-3.1, muse-image-1.0, muse-voice-transcribe-1.0) deliberately not added. variantMaps are NOT used: config-pinned expansion ignores them (developing-today-forks/opencode#540). Tiers address pools only at natively wired rungs until that lands.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 42 minutes. View limit details
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 |
This was referenced Oct 26, 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
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.
Implements the 2026-10-26 pool effort audit (all ladders verified against the serve-resolved model list; capability figures are Artificial Analysis Intelligence Index v4.3.2).
What changed (
.opencode/opencode.jsonconly)#low.#mediumsilently pruned every ClinePass leaf (its ladders have no medium; GLM has no medium on any door — Z.ai's API rejects it), so scout flash/GLM work ran OpenDesign-only, andhaiku-plancarried no selection at all.#lowis the only rung wired on every leaf of every door; Flash (max 45) and GLM (max 42) are above this tier's band at any rung.meta/muse-spark-1.3→#low(was bare). A 25-call probe against the Meta API showed low is the highest Spark rung that separates from the default plateau (mean reasoning tokens 644 vs 1,062 default; high ≈ default); quality was 5/5 at every rung on the test task.#high→#xhigh(51);muse-wonder-1.0→#max(52, was bare). Tier now reads 54 / 51 / 52.highcarried settings; selecting low/medium/xhigh/max recorded the choice and sent a default request.-lowmodel ids behind this door (probed: 404); effort is the reasoning_effort variants.Deliberately not in this PR
Per-leaf
variantMapremaps (the audit's Job 1 table). Route expansion prunes leaves on native variant names only and never consults the map for config-pinned variants — maps would be inert config. Filed as developing-today-forks/opencode#540; tiers instead address pools only at natively wired rungs until that lands.Validation
JSON.parseof the full file).validateModelRoutesreturns no refusals on the edited routes (harness contrast-checked against a deliberately duplicated target, which it does refuse).