fix(zai): expose GLM-5.2 reasoning levels [AI-assisted] by BorClaw · Pull Request #94136 · openclaw/openclaw

BorClaw · 2026-06-17T12:27:20Z

Summary

Problem: Z.AI glm-5.2 was present in the bundled provider, but OpenClaw still exposed Z.AI thinking as binary off/on and did not send Z.AI reasoning_effort for GLM-5.2 high/max thinking.
Why it matters: GLM-5.2 users could not select the provider's full off, low, high, max reasoning controls from OpenClaw.
What changed: glm-5.2 now advertises off, low, high, and max; payload wrapping maps low/high to reasoning_effort: "high", max/xhigh to reasoning_effort: "max", and off/minimal to thinking: { type: "disabled" }.
What did NOT change (scope boundary): Older Z.AI GLM models keep binary off/on; GLM-5.2 catalog metadata was already correct upstream and is not changed here.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes #
Related feat(providers): add GLM-5.2 support #92796

User-visible / Behavior Changes

Users selecting zai/glm-5.2 can choose off, low, high, or max thinking levels. Other Z.AI GLM models continue to show binary off/on thinking controls.

Security Impact (required)

New permissions/capabilities? (Yes/No) No
Secrets/tokens handling changed? (Yes/No) No
New/changed network calls? (Yes/No) No
Command/tool execution surface changed? (Yes/No) No
Data access scope changed? (Yes/No) No
If any Yes, explain risk + mitigation: N/A

Repro + Verification

Environment

OS: macOS
Runtime/container: local Node/pnpm checkout
Model/provider: zai/glm-5.2
Integration/channel (if any): bundled Z.AI provider plugin
Relevant config (redacted): saved Z.AI auth/config; no API key printed

Steps

Resolve the Z.AI provider thinking profile for glm-5.2 and glm-5.1.
Wrap a Z.AI glm-5.2 stream with low, high, and max thinking levels.
Verify emitted payload fields and docs.
Run live one-shot GLM-5.2 provider calls through this branch's openclaw infer model run path with off, high, and max thinking.

Expected

glm-5.2 exposes off, low, high, max.
glm-5.1 remains binary off/on.
low and high send reasoning_effort: "high".
max sends reasoning_effort: "max".
off disables thinking.
Live GLM-5.2 calls with off, high, and max return successful model text.

Actual

Matches expected in focused tests and live Z.AI GLM-5.2 probes.

Evidence

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Focused coverage was added in extensions/zai/index.test.ts for profile exposure and payload mapping.

Live redacted proof from this PR branch:

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking off --prompt 'Reply exactly: ZAI_off_OK' --json
{"ok":true,"capability":"model.run","transport":"local","provider":"zai","model":"glm-5.2","outputs":[{"text":"ZAI_off_OK"}]}

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking high --prompt 'Reply exactly: ZAI_high_OK' --json
{"ok":true,"capability":"model.run","transport":"local","provider":"zai","model":"glm-5.2","outputs":[{"text":"ZAI_high_OK"}]}

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking max --prompt 'Reply exactly: ZAI_max_OK' --json
{"ok":true,"capability":"model.run","transport":"local","provider":"zai","model":"glm-5.2","outputs":[{"text":"ZAI_max_OK"}]}

No API key or auth material is included in the proof output.

Human Verification (required)

Verified scenarios:
- CI=true pnpm format -- docs/tools/thinking.md docs/providers/zai.md extensions/zai/index.ts extensions/zai/index.test.ts
- CI=true pnpm exec vitest run extensions/zai/index.test.ts extensions/zai/model-definitions.test.ts src/auto-reply/thinking.test.ts
- CI=true pnpm tsgo
- CI=true pnpm lint
- Live: pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking off --prompt 'Reply exactly: ZAI_off_OK' --json
- Live: pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking high --prompt 'Reply exactly: ZAI_high_OK' --json
- Live: pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking max --prompt 'Reply exactly: ZAI_max_OK' --json
Edge cases checked:
- glm-5.1 remains binary off/on.
- GLM-5.2 max uses the new native max thinking level rather than the older xhigh workaround.
What you did not verify:
- Live provider behavior for every alias (low, medium, adaptive, xhigh); the live proof covers off, high, and max, while tests cover the alias mapping.

AI-assisted: yes. Patch authored with Codex/OpenClaw assistance and manually reviewed before submission.

Review Conversations

I replied to or resolved every bot review conversation I addressed in this PR.
I left unresolved only the conversations that still need reviewer or maintainer judgment.

If a bot review conversation is addressed by this PR, resolve that conversation yourself. Do not leave bot review conversation cleanup for maintainers.

Compatibility / Migration

Backward compatible? (Yes/No) Yes
Config/env changes? (Yes/No) No
Migration needed? (Yes/No) No
If yes, exact upgrade steps: N/A

Failure Recovery (if this breaks)

How to disable/revert this change quickly: revert this PR.
Files/config to restore: extensions/zai/index.ts, extensions/zai/index.test.ts, docs/providers/zai.md, docs/tools/thinking.md.
Known bad symptoms reviewers should watch for: Z.AI rejecting reasoning_effort, incorrect /think choices for GLM-5.2, or unexpected changes to non-GLM-5.2 Z.AI thinking choices.

Risks and Mitigations

Risk: Z.AI may change accepted reasoning_effort values.
- Mitigation: mapping is isolated to the Z.AI provider plugin, covered by tests, and live-smoked against GLM-5.2 for off, high, and max.

clawsweeper · 2026-06-17T12:57:36Z

Codex review: needs maintainer review before merge. Reviewed June 20, 2026, 11:56 PM ET / 03:56 UTC.

Summary
The PR expands Z.AI GLM-5.2 thinking options, maps selected levels to outbound reasoning_effort, and updates focused provider tests plus provider/thinking docs.

PR surface: Source +39, Tests +102, Docs +5. Total +146 across 4 files.

Reproducibility: yes. source-level. Current main exposes only binary Z.AI thinking and lacks GLM-5.2 reasoning_effort; the PR body adds after-fix live CLI output for off, high, and max.

Review metrics: 2 noteworthy metrics.

Provider thinking profile: 1 model expanded from 2 to 4 levels. This changes user-visible /think choices for zai/glm-5.2 while preserving binary choices for older Z.AI GLM models.
Outbound provider payload: 1 model-specific field added. reasoning_effort is an external Z.AI request contract maintainers should notice before merge.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🦞 diamond lobster
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

none.

Risk before merge

[P1] Merging changes the outbound Z.AI GLM-5.2 request contract by adding reasoning_effort; current Z.AI docs and contributor live output support it, but provider contract drift would be user-visible.
[P1] GitHub currently reports an unstable merge state with a failed Real behavior proof check despite sufficient body proof, so required checks still need normal maintainer handling.

Maintainer options:

Accept the GLM-5.2 provider contract (recommended)
Maintainers can merge after accepting the GLM-5.2-specific Z.AI reasoning_effort contract because the code path is isolated to the Z.AI provider hook and backed by docs, tests, and live CLI output.
Ask for one more low-level smoke
If maintainers want tighter live proof, ask for a redacted --thinking low run to directly cover the low-level mapping to Z.AI high effort.
Pause if Z.AI contract drifts
If current Z.AI docs or credentials stop accepting reasoning_effort, pause until runtime mapping and docs can be updated together.

Next step before merge

[P2] The remaining action is maintainer acceptance of the Z.AI provider-contract risk and normal required-check handling, not an automated code repair.

Security
Cleared: The diff changes Z.AI provider payload mapping, focused tests, and docs only; it does not add dependencies, workflows, permissions, credential handling, or new code execution paths.

Review details

Best possible solution:

Land the provider-owned GLM-5.2 thinking profile and payload mapping after maintainers accept the Z.AI request-contract risk and required checks are green or explicitly handled.

Do we have a high-confidence way to reproduce the issue?

Yes, source-level. Current main exposes only binary Z.AI thinking and lacks GLM-5.2 reasoning_effort; the PR body adds after-fix live CLI output for off, high, and max.

Is this the best way to solve the issue?

Yes. The provider-owned wrapStreamFn and resolveThinkingProfile hooks are the narrow owner boundary for a Z.AI-only model request contract, with no new core/config surface.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 6c1041339da6.

Label changes

Label justifications:

P2: This is a focused provider runtime/docs improvement with limited blast radius but real user-visible Z.AI behavior impact.
merge-risk: 🚨 auth-provider: The PR changes Z.AI model-specific thinking options and outbound request payloads for GLM-5.2, which depends on provider contract behavior.
rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🦞 diamond lobster and patch quality is 🐚 platinum hermit.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (live_output): The PR body and follow-up comment include redacted live CLI output showing successful after-fix GLM-5.2 calls for off, high, and max.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body and follow-up comment include redacted live CLI output showing successful after-fix GLM-5.2 calls for off, high, and max.

Evidence reviewed

PR surface:

Source +39, Tests +102, Docs +5. Total +146 across 4 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	47	8	+39
Tests	1	102	0	+102
Docs	2	6	1	+5
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	4	155	9	+146

What I checked:

Current main Z.AI binary thinking: Current main returns the tool-stream wrapper unchanged for non-off Z.AI thinking unless preserved thinking is set, and the registered thinking profile exposes only off plus binary on. (extensions/zai/index.ts:131, 6c1041339da6)
PR head runtime mapping: At PR head, GLM-5.2 maps low/medium/high/adaptive to reasoning_effort: "high", maps xhigh/max to "max", and keeps older GLM models binary. (extensions/zai/index.ts:131, 432214158a39)
PR head focused tests: At PR head, focused tests cover GLM-5.2 profile exposure, older GLM binary preservation, minimal-thinking preservation, and GLM-5.2 reasoning-effort payload mapping. (extensions/zai/index.test.ts:283, 432214158a39)
Provider SDK owner boundary: The provider SDK documents wrapStreamFn as provider-owned request-body rewrites and resolveThinkingProfile as the model-specific /think option set. (src/plugins/types.ts:1477, 6c1041339da6)
Gateway thinking profile consumer: Gateway session defaults use listThinkingLevelOptions from the selected provider/model profile, so the Z.AI provider hook drives UI options and validation. (src/gateway/session-utils.ts:1302, 6c1041339da6)
Z.AI contract proof: Official Z.AI Markdown docs describe reasoning_effort as supported for GLM-5.2 and above, list allowed values, and state low/medium map to high while xhigh maps to max.

Likely related people:

steipete: Authored and merged the related GLM-5.2 support PR and authored the latest branch commit preserving minimal-thinking semantics. (role: recent feature owner; confidence: high; commits: 735f59af7366, 432214158a39; files: extensions/zai/index.ts, extensions/zai/index.test.ts, extensions/zai/model-definitions.ts)
Vincent Koc: Current main blame for the Z.AI thinking wrapper/profile and thinking docs points to this author, and nearby history shows provider stream/default hook work. (role: recent area contributor; confidence: high; commits: 3f166b1f6411, 3968fea383b2, 9c42e6424d25; files: extensions/zai/index.ts, docs/tools/thinking.md, docs/providers/zai.md)
huangjianxiong: Recently touched synthesized GLM-5 Z.AI model base URL behavior adjacent to the same provider/model family. (role: recent adjacent contributor; confidence: medium; commits: db54a3268b77; files: extensions/zai/index.ts, extensions/zai/model-definitions.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

BorClaw · 2026-06-17T13:14:37Z

Updated the PR with the two missing pieces ClawSweeper called out:

docs/tools/thinking.md now documents the GLM-5.2 exception for Z.AI thinking levels.
Added redacted live GLM-5.2 proof to the PR body.

Live proof run from this PR branch, with no API key or auth material printed:

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking off --prompt 'Reply exactly: ZAI_off_OK' --json
ok=true provider=zai model=glm-5.2 output=ZAI_off_OK

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking high --prompt 'Reply exactly: ZAI_high_OK' --json
ok=true provider=zai model=glm-5.2 output=ZAI_high_OK

$ pnpm openclaw infer model run --local --model zai/glm-5.2 --thinking max --prompt 'Reply exactly: ZAI_max_OK' --json
ok=true provider=zai model=glm-5.2 output=ZAI_max_OK

Local verification also passed:

CI=true pnpm format -- docs/tools/thinking.md docs/providers/zai.md extensions/zai/index.ts extensions/zai/index.test.ts
CI=true pnpm exec vitest run extensions/zai/index.test.ts extensions/zai/model-definitions.test.ts src/auto-reply/thinking.test.ts
CI=true pnpm tsgo
CI=true pnpm lint

@clawsweeper re-review

clawsweeper · 2026-06-17T13:14:41Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

State: Failed
Detail: The targeted re-review timed out or failed while waiting for exact-review Codex capacity. Retry after capacity recovers.
Run: https://github.com/openclaw/clawsweeper/actions/runs/27691704400
Updated: 2026-06-17T13:37:25.092Z

steipete · 2026-06-21T03:57:53Z

Merged via squash.

Prepared head SHA: 432214158a39f6b90a89e71c0e8dd844b1195b5a
Merge commit: 50c2cc6a45642cfcb98e3fe174873fc93a255c48

Thanks @BorClaw!

@steipete

Merged via squash. Prepared head SHA: 4322141 Co-authored-by: BorClaw <[email protected]> Co-authored-by: steipete <[email protected]> Reviewed-by: @steipete

openclaw-barnacle Bot added docs Improvements or additions to documentation agents Agent runtime and tooling size: S triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels Jun 17, 2026

BorClaw force-pushed the fix/zai-glm-thinking-levels branch from b5c21a9 to 89298df Compare June 17, 2026 12:34

openclaw-barnacle Bot added extensions: zai and removed agents Agent runtime and tooling labels Jun 17, 2026

BorClaw changed the title ~~fix(zai): expose GLM-5 reasoning levels [AI-assisted]~~ fix(zai): expose GLM-5.2 reasoning levels [AI-assisted] Jun 17, 2026

openclaw-barnacle Bot added the triage: needs-pr-context Candidate: external PR body lacks required problem context or evidence. label Jun 19, 2026

steipete force-pushed the fix/zai-glm-thinking-levels branch 2 times, most recently from e8a5c03 to 3e59875 Compare June 21, 2026 03:41

BorClaw and others added 3 commits June 20, 2026 23:49

fix(zai): expose GLM-5.2 reasoning levels

fe60665

docs: align Z.AI GLM-5.2 thinking guide

6eed4c2

fix(zai): preserve minimal thinking semantics

4322141

steipete force-pushed the fix/zai-glm-thinking-levels branch from 3e59875 to 4322141 Compare June 21, 2026 03:49

steipete merged commit 50c2cc6 into openclaw:main Jun 21, 2026
149 of 150 checks passed

github-actions Bot mentioned this pull request Jun 29, 2026

chore(openclaw): bump pinned version v2026.5.28 → v2026.6.10 Evans-Software-Solutions-Limited/axel-saas#120

Open

4 tasks

100yenadmin mentioned this pull request Jul 11, 2026

[Bug]: Gateway ignores Z.AI GLM-5.2 max thinking profile while local infer succeeds #104716

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(zai): expose GLM-5.2 reasoning levels [AI-assisted]#94136

fix(zai): expose GLM-5.2 reasoning levels [AI-assisted]#94136
steipete merged 3 commits into
openclaw:mainfrom
BorClaw:fix/zai-glm-thinking-levels

BorClaw commented Jun 17, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading

Uh oh!

BorClaw commented Jun 17, 2026

Uh oh!

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading

Uh oh!

Uh oh!

steipete commented Jun 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

BorClaw commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Review Conversations

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

clawsweeper Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BorClaw commented Jun 17, 2026

Uh oh!

clawsweeper Bot commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

steipete commented Jun 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

BorClaw commented Jun 17, 2026 •

edited

Loading

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading