Repository navigation
fix(providers): keep grok-4.7 reasoning tiers from object catalogs - #14525
Merged
diegosouzapw merged 5 commits intoSep 25, 2026
Merged
diegosouzapw merged 5 commits into
diegosouzapw merged 5 commits into
Conversation
This was referenced Sep 23, 2026
Collaborator
|
Nice catch — the live-catalog check for grok-4.7/4.7-build-fast/4.6/4.5 makes the tier list |
4 tasks done
HouMinXi
force-pushed
the
fix/grok-build-object-efforts
branch
from
September 24, 2026 15:16
84c998a to
e001ff6
Compare
Grok Build /v1/models now returns reasoning_efforts as objects
({id, value, label, default}). The discovery parser kept only bare
strings, so the list collapsed to empty and the UI had no tiers to
set. grok-4.7 was also missing from the seed, and the executor only
filled the default high effort for grok-4.5 and grok-4.6.
Read value or id off each object, accept xhigh when the payload names
it, seed grok-4.7 with the four live tiers, and apply the default
effort to any non-composer model that omits one. A payload with no
tier list still falls back to low/medium/high, because grok-4.5 does
not advertise xhigh.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
An object tier with value "" still counted as a string, so a usable id on the same object was dropped. Trim the value first and fall through to id when nothing remains. Signed-off-by: Minxi Hou <houminxi@gmail.com>
Signed-off-by: Minxi Hou <houminxi@gmail.com>
…ations The flagship guard test and the discovery-split guard now run against a registry where grok-4.7 is seeded on purpose and object catalogs carry xhigh, so the expectations follow the new semantics instead of banning the model this change adds. Signed-off-by: Minxi Hou <houminxi@gmail.com>
HouMinXi
force-pushed
the
fix/grok-build-object-efforts
branch
from
September 24, 2026 18:47
e001ff6 to
73e93c9
Compare
diegosouzapw
pushed a commit
that referenced
this pull request
Sep 25, 2026
…attempts (#14474) Merged in the 2026-09-25 maintainer merge-batch — thank you for the contribution! The maintainer rework applied to this branch (if any) is described in the "Maintainer rework (merge-batch)" section of the PR body. Re-validated in a combined tree with the other ready Track-B PRs (#14474, #14188, #14525, #14807, #14796) on the current `release/v3.8.51` tip: - 182/183 focused node tests. The one red (`combo-skipped-reset-timing`) passes 1/1 when run on its own; it is a timing flake under devbox load 95. - `typecheck:core`: 0 errors; the tests these PRs previously broke (video-bridge-log-redaction, quota-reset-timing, flagship-0day-discovery, provider-models-discovery-split) are all green in the combined tree.
diegosouzapw
merged commit Sep 25, 2026
8e6d935
into
cheaper-inference:release/v3.8.51
11 of 16 checks passed
diegosouzapw
pushed a commit
that referenced
this pull request
Sep 25, 2026
Merged in the 2026-09-25 maintainer merge-batch — thank you for the contribution! The maintainer rework applied to this branch (if any) is described in the "Maintainer rework (merge-batch)" section of the PR body. Re-validated in a combined tree with the other ready Track-B PRs (#14474, #14188, #14525, #14807, #14796) on the current `release/v3.8.51` tip: - 182/183 focused node tests. The one red (`combo-skipped-reset-timing`) passes 1/1 when run on its own; it is a timing flake under devbox load 95. - `typecheck:core`: 0 errors; the tests these PRs previously broke (video-bridge-log-redaction, quota-reset-timing, flagship-0day-discovery, provider-models-discovery-split) are all green in the combined tree.
diegosouzapw
pushed a commit
that referenced
this pull request
Sep 25, 2026
Merged in the 2026-09-25 maintainer merge-batch — thank you for the contribution! The maintainer rework applied to this branch (if any) is described in the "Maintainer rework (merge-batch)" section of the PR body. Re-validated in a combined tree with the other ready Track-B PRs (#14474, #14188, #14525, #14807, #14796) on the current `release/v3.8.51` tip: - 182/183 focused node tests. The one red (`combo-skipped-reset-timing`) passes 1/1 when run on its own; it is a timing flake under devbox load 95. - `typecheck:core`: 0 errors; the tests these PRs previously broke (video-bridge-log-redaction, quota-reset-timing, flagship-0day-discovery, provider-models-discovery-split) are all green in the combined tree.
3 of 6 tasks
deptrai
pushed a commit
to deptrai/OmniRoute
that referenced
this pull request
Sep 28, 2026
…heaper-inference#14525) Merged in the 2026-09-25 maintainer merge-batch — thank you for the contribution! The maintainer rework applied to this branch (if any) is described in the "Maintainer rework (merge-batch)" section of the PR body. Re-validated in a combined tree with the other ready Track-B PRs (cheaper-inference#14474, cheaper-inference#14188, cheaper-inference#14525, cheaper-inference#14807, cheaper-inference#14796) on the current `release/v3.8.51` tip: - 182/183 focused node tests. The one red (`combo-skipped-reset-timing`) passes 1/1 when run on its own; it is a timing flake under devbox load 95. - `typecheck:core`: 0 errors; the tests these PRs previously broke (video-bridge-log-redaction, quota-reset-timing, flagship-0day-discovery, provider-models-discovery-split) are all green in the combined tree.
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.
Summary
Grok Build
/v1/modelsnow returns each reasoning tier as an object (id,value,label,default) instead of a bare string. Discovery kept only strings, so a non-empty object list counted as explicit and collapsed to nothing. The settings UI then had no tiers for grok-4.7. The built-in seed also stopped at grok-4.6, and the executor filled the defaulthigheffort only for grok-4.5 and grok-4.6.What the live catalog says
Checked a live Grok Build catalog on 2026-09-22. grok-4.7, grok-4.7-build-fast, and grok-4.6 advertise
xhigh,high,medium,low, with defaulthigh. grok-4.5 advertiseshigh,medium,lowand has noxhigh.Changes
value, thenid, from each tier object. A blankvalueno longer hides a usableid.xhighonly when the payload names it. A catalog that omits the list still falls back to low/medium/high, so grok-4.5 does not grow a tier it does not advertise.xhighin the seed as well.higheffort to any non-composer model that omits one, including grok-4.7. Composer still drops the effort.Validation
node --import tsx/esm --test --test-force-exit tests/unit/grok-build-object-efforts.test.ts tests/unit/grok-cli-responses-compat.test.ts tests/unit/grok-cli-reasoning-strip-6288.test.ts20/20 passing. The empty-value case fails if the parser treats
value: ""as a chosen tier.Maintainer rework (merge-batch 2026-09-24)
release/v3.8.51into the branch (clean). No code changes from the maintainer.internal/registry/models/models.json(as of 2026-09-25) lists Grok Buildgrok-4.7,grok-4.7-build-fastandgrok-4.6withlow/medium/high/xhigh, andgrok-4.5withlow/medium/highonly. So the updated expectations inprovider-models-discovery-split.test.tsandflagship-0day-discovery.test.tstest 9 (73e93c9) are corrections, not weakened tests. The guard still bans the other unreleased ids.