Skip to content

fix(providers): restore glm-5.3 to both Alibaba Token Plan catalogs - #5010

Merged
lidge-jun merged 1 commit into
lidge-jun:devfrom
oliver-mee:fix/token-plan-glm-53
Sep 18, 2026
Merged

lidge-jun merged 1 commit into
lidge-jun:devfrom
oliver-mee:fix/token-plan-glm-53

Conversation

@oliver-mee

@oliver-mee oliver-mee commented Sep 18, 2026

Copy link
Copy Markdown
Contributor

Summary

  • glm-5.3 returns to both Alibaba Token Plan catalogs (alibaba-token-plan Beijing and alibaba-token-plan-intl Singapore). The 260909 refresh removed it as a phantom, which was correct at the time: the id 404'd on every plan key. The gateway started serving it on 260917: /models now lists glm-5.3 for global Team, global Personal, and CN Team, and it answers completions on a Personal key (probed 260918).
  • Measured contract on the plan gateway (probed 2026-09-18, evidence in the dated comments): effort ladder low|high|max (reuses the existing ZAI_GLM_53_REASONING_EFFORTS const), thinking always-on (the gateway rejects enable_thinking:false with 400 restricted to True, unlike glm-5.2), 1M context, 131,072 max output, text-only input, strict json_schema accepted so no noJsonSchemaModels wiring needed.
  • glm-5.3-flash stays out of both rosters: it is still never served by the Token Plan gateway (Z.ai VLM id, not plan entitlement). The parity golden test and the catalog assertions are updated in this same commit, per the repo convention that catalog PRs keep golden parity current.

Verification

  • bun test tests/gui/alibaba-intl-token-plan.test.ts - 16 pass, 0 fail.
  • bun test tests/providers/provider-registry-parity.test.ts - 53 pass, 0 fail.
  • bun test tests/claude-integration/claude-desktop-1m.test.ts - 2 pass, 0 fail.
  • bun run typecheck - clean.
  • bun run privacy:scan - passed.
  • git diff --check - clean.
  • Full suite not run (low-RAM VPS makes broad bun test lanes unreliable; focused lanes above cover every file this diff touches).

AI-assisted: prepared with Hermes agent on behalf of the account owner; catalogue facts and the gateway probes above were performed and dated by the maintainer of https://github.com/oliver-mee/alibaba-token-plan-wiki (this repo's cited evidence source for Token Plan rows).

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

  • All CI tests are green on my local testing.

  • I pushed my PR to the latest dev commit.

  • I resolved all correct Codex and CodeRabbit findings.

  • My PR is ready for review.

The 260909 refresh removed glm-5.3 as a phantom (the gateway 404'd the id
then). It is real now: /models lists it for global Team, global Personal,
and CN Team from 260917, and it answers completions on a Personal key
(probed 260918). Contract on the plan gateway: effort low/high/max
(default max), thinking always-on (enable_thinking:false rejected with
400), 1M context, 131,072 max output, text-only input, strict json_schema
accepted. glm-5.3-flash stays out: still never served by the plan.

Evidence: https://github.com/oliver-mee/alibaba-token-plan-wiki
(machine-readable catalogue, data/models.json, updated 2026-09-18).
@github-actions

Copy link
Copy Markdown
Contributor

Deterministic PR hygiene checks passed.

@github-actions github-actions Bot added the bug Something isn't working label Sep 18, 2026
@github-actions

github-actions Bot commented Sep 18, 2026

Copy link
Copy Markdown
Contributor

✅ READY

  • all PR quality gates passed; the review readiness checklist is complete.

Review readiness checklist

  • ✅ All CI tests are green on my local testing.
  • ✅ I pushed my PR to the latest dev commit.
  • ✅ I resolved all correct Codex and CodeRabbit findings.
  • ✅ My PR is ready for review.

4/4 boxes ticked.

This pull request is already Ready for Review.
The review-ready label marks this PR as ready; review automation runs independently.
Maintainers: @lidge-jun @Ingwannu

@github-actions
github-actions Bot marked this pull request as draft September 18, 2026 03:03
@coderabbitai

coderabbitai Bot commented Sep 18, 2026

Copy link
Copy Markdown
Contributor

Review Change StackReview Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 1ee89fa6-107d-45dc-8056-3771c5f886e4

📥 Commits

Reviewing files that changed from the base of the PR and between c9fd114 and b3d954d.

📒 Files selected for processing (4)
  • src/providers/registry/entries-extended.ts
  • src/providers/registry/model-seeds.ts
  • tests/gui/alibaba-intl-token-plan.test.ts
  • tests/providers/provider-registry-parity.test.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.


📝 Walkthrough

Walkthrough

GLM-5.3 is restored to the Beijing and International Alibaba Token Plan registries with its model metadata. Provider roster and parity tests now expect GLM-5.3 and continue to exclude GLM-5.3-flash.

Changes

Alibaba GLM-5.3 restoration

Layer / File(s) Summary
Restore GLM-5.3 registry metadata
src/providers/registry/model-seeds.ts, src/providers/registry/entries-extended.ts
Adds GLM-5.3 to both Token Plan model lists. Sets text-only input, a 1,000,000-token context window, 131,072 maximum output tokens, no-vision support, preserved reasoning, and low/high/max reasoning efforts.
Update provider roster validation
tests/gui/alibaba-intl-token-plan.test.ts, tests/providers/provider-registry-parity.test.ts
Updates Beijing and International roster expectations to include GLM-5.3. The tests continue to exclude GLM-5.3-flash and verify GLM-5.3 metadata.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to b3d95

The restored model catalog matches the intended provider metadata and roster behavior, so no merge-blocking risk remains.

🚥 Pre-merge checks | ✅ 5
✅ Passed checks (5 passed)
Check name Status Explanation
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 4…
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
Description Check ✅ Passed Check skipped - CodeRabbit’s high-level summary is enabled.
Title check ✅ Passed The title clearly and concisely describes the main change: restoring glm-5.3 to both Alibaba Token Plan catalogs.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create a new PR

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.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@github-actions
github-actions Bot marked this pull request as ready for review September 18, 2026 03:05
@lidge-jun

Copy link
Copy Markdown
Owner

리뷰 · 우선순위 74 / 80

이 PR은 Alibaba Token Plan 두 카탈로그에서 빠진 glm-5.3 한 줄을 다시 넣는 수정이다. 지금 dev HEAD c9fd114cfac61570f249886a954800bd7dfa689e (패키지 2.59.0, tip #4987 Dashboard Fast row) 를 보면 src/providers/registry/model-seeds.tsALIBABA_TOKEN_PLAN_MODELS (베이징 Personal) 와 ALIBABA_INTL_TOKEN_PLAN_MODELS (싱가포르 Team) 둘 다 glm-5.2 까지만 있고 glm-5.3 은 없다. 주석에도 260909 게이트웨이 재프로브 때 Token Plan이 이 id를 안 줘서 팬텀으로 지웠다고 적혀 있다. 같은 파일의 Z.AI Coding Plan·Volcengine Plan·Zhipu 쪽에는 이미 glm-5.3 이 있으니, Token Plan만 뒤처진 상태다.

작성자 설명은 이렇게 이어진다. 260909 삭제는 그때 기준으로 맞았다(모든 plan 키에서 404). 그런데 260917부터 게이트웨이 /modelsglm-5.3 이 global Team·global Personal·CN Team에 뜨고, 260918 Personal 키로 completion도 통과했다. 증거는 이 리포가 이미 Token Plan 행의 근거로 쓰는 https://github.com/oliver-mee/alibaba-token-plan-wiki 쪽이다. 그래서 이 PR은 카탈로그만 다시 맞추는 일이지, 새 프로바이더를 만드는 일이 아니다.

코드 변경은 네 파일로 좁다. model-seeds.ts 에서 베이징·인터내셔널 모델 배열에 glm-5.3 을 넣고, ALIBABA_TOKEN_PLAN_INPUT_MODALITIES["text"], 공유 맵 ALIBABA_TOKEN_PLAN_CONTEXT_WINDOWS / ALIBABA_TOKEN_PLAN_MAX_OUTPUT_TOKENS 에 1_000_000 / 131_072, ALIBABA_TOKEN_PLAN_NO_VISIONALIBABA_TOKEN_PLAN_PRESERVE_REASONING 에도 같은 id를 넣는다. glm-5.3-flash 는 일부러 넣지 않는다(Token Plan이 여전히 안 줌). entries-extended.tsalibaba-token-plan / alibaba-token-plan-intl modelReasoningEfforts 에는 이미 있는 ZAI_GLM_53_REASONING_EFFORTS (low|high|max) 를 연결한다. 테스트는 tests/gui/alibaba-intl-token-plan.test.ts 가 not-contain → contain으로 바뀌고 길이 19→20, 노력·컨텍스트·출력·모달리티·preserve 단언이 붙으며, tests/providers/provider-registry-parity.test.ts 베이징 golden 모델 목록도 같이 맞춰진다. 카탈로그 PR이 golden 패리티를 같이 고치는 관례와 같다.

현재 스냅샷 방향(프로바이더/카탈로그 충실도, 릴리스 규율)과도 잘 맞는다. types.ts/config.ts 분할 캠페인에 걸려 무효화될 종류의 패치가 아니다. 닫고 리베이스 말라는 케이스가 아니다. 다만 PR은 아직 draft이고 readiness 체크리스트가 비어 있으며, 브랜치 부모는 444cf77012a6 (#4986 SOCKS5) 이라 tip c9fd114c 대비 약 10커밋 뒤처져 diverged다. 머지 전에 tip 위로 리베이스가 필요하다.

한 가지 계약 구멍도 있다. PR 본문은 plan 게이트웨이에서 effort 기본값이 max이고 thinking이 always-on(enable_thinking:false → 400)이라고 적었는데, 두 Token Plan 엔트리의 modelDefaultReasoningEfforts 는 여전히 QWEN38_FAMILYxhigh 만 있고 glm-5.3max 가 없다. 같은 파일에서 Z.AI Coding Plan 쪽은 ZAI_GLM_53_MODELS 기본값을 max로 이미 박아 두었다. Token Plan만 기본값이 비면 픽커/시드가 프로브와 어긋날 수 있다. always-on thinking을 레지스트리에 어떻게 고정할지도(토글 노출 vs 어댑터에서 false 억제) 아직 코드에 반영되지 않았다.

라인 825/874 (entries-extended.ts alibaba-token-plan / alibaba-token-plan-intl modelDefaultReasoningEfforts) - 프로브는 glm-5.3 기본 effort=max인데 Token Plan 엔트리에 기본값이 안 붙어 Qwen3.8 기본만 남는다
경로 PR draft + readiness checklist - 네 칸이 모두 비어 있고 isDraft=true라 mergeStateStatus=BLOCKED다. 레디 표시 전에는 랜딩 대상이 아니다
경로 head 부모 444cf77 vs tip c9fd114 - ahead 1 / behind 10 diverged. tip 위 리베이스 없이 깔끔한 랜딩이 어렵다
경로 thinking always-on 계약 - 본문은 enable_thinking:false 400을 말했지만 Token Plan 엔트리에 thinkingToggle 고정/억제가 없다. UI가 false를내면 런타임 400이 날 수 있다
경로 provider-registry-parity 베이징 golden - 모델 목록만 갱신되고 effort/context/maxOutput/modality 단언은 intl GUI 테스트 쪽에만 있다. 패리티 대칭이 약하다
경로 glm-5.3-flash - 의도적으로 제외한 선택은 맞다. 다시 넣지 말 것

메인테이너의 판단이 필요한 지점

  • Token Plan glm-5.3에 modelDefaultReasoningEfforts max를 같은 커밋에 넣을지, 후속 초소형 PR로 manifest는지
  • thinking always-on을 레지스트리/어댑터에서 강제할지, 주석·위키 증거만으로 둘지
  • draft 체크리스트를 작성자가 채운 뒤 Ready로 올릴지, 메인테이너가 증거 신뢰하고 바로 Ready 처리할지
  • 베이징 Personal 자격 주장을 위키 data/models.json 한 번 더 눈으로 확인할지(작성자가 위키 메인테이너이기도 함)

너의 추천
병합 후보로 둔다. 닫지 말고, tip c9fd114c 위로 리베이스한 뒤 entries-extended.ts 두 Token Plan 엔트리에 modelDefaultReasoningEfforts["glm-5.3"] = "max" 를 추가하고(가능하면 always-on 처리도 한 줄로 명확히), Ready 체크리스트를 채운 다음 focused 테스트(alibaba-intl-token-plan + provider-registry-parity) 초록을 확인한 뒤 랜딩한다. types/config 분할에 무효화되지 않는다. glm-5.3-flash 는 계속 제외.

이 댓글은 grok-bot이 작성했습니다

@lidge-jun
lidge-jun merged commit 6fc8289 into lidge-jun:dev Sep 18, 2026
12 of 13 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working review-ready

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants