Repository navigation
fix(providers): drop stale zhipuai-coding-plan entries after GLM-5.3 consolidation - #7331
Merged
rekram1-node merged 1 commit intoSep 17, 2026
Conversation
…consolidation Zhipu's GLM Coding Plan consolidated to GLM-5.3 and GLM-5.3-Flash (official announcement: docs.bigmodel.cn/cn/coding-plan/overview, "可用模型": all plans serve GLM-5.3 and GLM-5.3-Flash only). First-hand verification against the live coding-plan endpoint (open.bigmodel.cn/api/coding/paas/v4, current subscription): | catalog id | server answers | action | |--------------------|---------------------------|--------| | glm-5.3 | glm-5.3 | keep | | glm-5.3-flash | glm-5.3-flash | keep | | glm-4.6v | glm-4.6v (still served!) | keep | | glm-5.2 | glm-5.3 | drop | | glm-5.1 | glm-5.3 | drop | | glm-5-turbo | glm-5.3-flash | drop | | glm-4.7 | glm-5.3-flash | drop | | glm-4.5-air | glm-5.3-flash | drop | | glm-5.2-highspeed | 403 not in plan | drop | | glm-5.3-highspeed | 403 not in plan | drop | | glm-5v-turbo | 403 not in plan | drop | Keeping rerouted entries means a user picking e.g. glm-5-turbo talks to GLM-5.3-Flash while the catalog applies stale context/maxTokens/ cost/reasoning metadata to the session. Note: glm-4.6v is not mentioned in the announcement but is still served as-is on coding plans, so it stays — it would have been wrongly dropped by announcement-only triage. Mirrors the earlier deepseek consolidation. Refs earendil-works/pi#9616 (pi mirrors this provider; maintainer asked for a volunteer to sync the catalog upstream). bun validate passes; the assembled provider catalog now exposes exactly glm-5.3, glm-5.3-flash, glm-4.6v.
Contributor
|
No actionable findings. |
github-actions
Bot
added
the
reviewer: ready
heimoshuiyu
added a commit
that referenced
this pull request
Sep 17, 2026
#7331 classified glm-5.3-highspeed as stale from a single-plan 403 'not available on plan'. That is tier access control, not retirement: the coding endpoint still serves the id as-is on eligible plan tiers (first-hand verified 2026-09-17). The same PR kept the byte-identical zai-coding-plan twin, and tier-gated highspeed variants are established in this catalog (kimi-for-coding, minimax-coding-plan). Only glm-5.3-highspeed is restored; the other entries dropped in #7331 have no evidence of being served on any tier. Leading comment now documents the tier gating.
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.
What
Drop the 8 stale entries under
providers/zhipuai-coding-plan/models/, keepingglm-5.3,glm-5.3-flash, andglm-4.6v.Deleted:
glm-4.5-air,glm-4.7,glm-5-turbo,glm-5.1,glm-5.2,glm-5.2-highspeed,glm-5.3-highspeed,glm-5v-turbo.Why
Zhipu's GLM Coding Plan consolidated its lineup — per the official announcement (docs.bigmodel.cn/cn/coding-plan/overview, "可用模型"): all plans now serve GLM-5.3 and GLM-5.3-Flash only. Keeping stale entries means a user picking e.g.
glm-5-turboactually talks to GLM-5.3-Flash while downstream consumers (pi and anything else mirroring this provider) apply the stale entry's context window, maxTokens, cost, and reasoning metadata to the session.Mirrors the earlier deepseek consolidation.
Verification
First-hand check against the live coding-plan endpoint (
open.bigmodel.cn/api/coding/paas/v4/chat/completions, current subscription; responsemodelfield shown):glm-5.3glm-5.3glm-5.3-flashglm-5.3-flashglm-4.6vglm-4.6v(still served, despite not being in the announcement)glm-5.2glm-5.3glm-5.1glm-5.3glm-5-turboglm-5.3-flashglm-4.7glm-5.3-flashglm-4.5-airglm-5.3-flashglm-5.2-highspeedglm-5.3-highspeedglm-5v-turbobun validatepasses; the assembled provider catalog exposes exactlyglm-5.3,glm-5.3-flash,glm-4.6v.One finding worth flagging:
glm-4.6vis not mentioned in Zhipu's announcement but is still served as-is on coding plans — announcement-only triage would have wrongly dropped it.Context
Upstream consumer issue: earendil-works/pi#9616 — pi mirrors
zhipuai-coding-planfrom models.dev and currently exposes 10 entries, 6 of which silently route to different models. pi's maintainer asked there for a volunteer to sync the catalog upstream; this is that sync. Changes prepared with AI assistance, verified first-hand against the live endpoint by the (human) GLM Coding Plan subscriber.