fix: add GLM-5.2 to models supporting max reasoning effort

[is911]

Description

GLM-5.2 (Z.ai) natively supports reasoning_effort: "max" alongside "high" (Z.ai docs). However, normalizeOpenAIReasoningEffort() silently downgraded "max" to "high" for GLM-5.2 because it was not listed in supportsMaxReasoningEffort(). This is the same normalization gap that previously affected DeepSeek V4 (#4320, fixed in #4330).

Evidence: a direct probe through a Bifrost gateway (v1.5.13) routing to Z.ai GLM-5.2 confirmed that reasoning_effort: "max" sent to Bifrost was forwarded to Z.ai as reasoning_effort: "high" (verified via extra_fields.raw_request).

Type of Change

• Bug fix
• Feature
• Refactor
• Documentation
• Chore/CI

Affected Packages

• core/providers/openai/

Changes Made

• core/providers/openai/utils.go: Added glm-5.2 to supportsMaxReasoningEffort() so normalizeOpenAIReasoningEffort() returns "max" unchanged for GLM-5.2 models (both bare glm-5.2 and provider-prefixed zai/glm-5.2).
• core/providers/openai/chat_test.go: Extended TestToOpenAIChatRequest_NormalizesReasoningEffort and TestOpenAIChatRequest_FilterOpenAISpecificParameters_NormalizesReasoningEffort with GLM-5.2 test cases (bare and provider-prefixed).
• core/providers/openai/responses_test.go: Extended TestToOpenAIResponsesRequest_NormalizesReasoningEffort with GLM-5.2 test cases (bare and provider-prefixed).
• core/changelog.md: Added changelog entry.

Testing

go test ./core/providers/openai/... -run "NormalizesReasoningEffort|FilterOpenAISpecificParameters" -v -count=1

All tests pass, including the new GLM-5.2 cases:

• preserves max for glm-5.2 ✓
• preserves max for provider-prefixed glm-5.2 ✓

Regression cases for existing models remain unchanged:

• preserves max for deepseek-v4-pro/flash ✓
• maps max to xhigh for gpt-5.4 ✓
• maps max to high for gpt-5.1 ✓

Code formatted with gofmt. No new lint issues introduced.

Checklist

• Code follows the project style guidelines (make fmt)
• Self-reviewed the code
• Added tests for new behavior
• All tests pass (go test ./core/providers/openai/...)
• Changelog updated

Related Issues

Closes #4460

[CLAassistant]

All committers have signed the CLA.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 284b923b-bfa1-4ddf-a05b-5b09411ffc53

📥 Commits

Reviewing files that changed from the base of the PR and between 526254e and 4c56d14.

📒 Files selected for processing (3)

• core/changelog.md
• core/providers/openai/chat_test.go
• core/providers/openai/responses_test.go

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • GLM-5.2 (Z.ai) models now correctly preserve the “max” reasoning effort setting.
  • Improved OpenAI-compatible gateway behavior, including provider handling and better tool-call metadata/results ordering across supported gateways.

• New Features

  • Added cache diagnostics to help explain cache hits/misses, including first cache-prefix divergence reporting.

• Tests

  • Expanded/extended reasoning-effort normalization coverage for GLM-5.2, including plain and provider-prefixed model names, with proper provider registration during test runs.

Walkthrough

supportsMaxReasoningEffort now recognizes glm-5.2 model prefixes, a changelog entry was added, and reasoning-effort tests register the zai provider and verify "max" stays unchanged for glm-5.2 models.

Changes

GLM-5.2 max reasoning effort support

Layer / File(s)	Summary
supportsMaxReasoningEffort: add glm-5.2 prefix check and document core/providers/openai/utils.go, core/changelog.md	supportsMaxReasoningEffort now returns true for glm-5.2 prefixes in addition to deepseek-v4, and the changelog records the update.
Test coverage: zai provider registration and glm-5.2 cases core/providers/openai/chat_test.go, core/providers/openai/responses_test.go	The chat and responses reasoning-effort tests register zai and add cases for both glm-5.2 and zai/glm-5.2 to confirm "max" is preserved.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Suggested reviewers

• danpiths
• akshaydeo

Poem

🐇 GLM hopped in, asking for more,
max stayed max at the gateway door.
With a prefix tweak bright,
And tests set just right,
The rabbit applauds: "Reasoning restored!"

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately and concisely describes the main change: adding GLM-5.2 to the list of models supporting max reasoning effort.
Description check	✅ Passed	The PR description is comprehensive and follows the template structure with clear sections for Summary, Type of Change, Affected Packages, Changes Made, Testing, and Checklist.
Linked Issues check	✅ Passed	The PR fully addresses issue #4460 by adding GLM-5.2 to supportsMaxReasoningEffort() function and extending tests to verify the fix for both bare and provider-prefixed model names.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to the fix for issue #4460: updates to utils.go, test files, and changelog. No unrelated modifications are present.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 golangci-lint (2.12.2)

level=error msg="[linters_context] typechecking error: pattern ./...: directory prefix . does not contain main module or its selected dependencies"

---

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

• X
• Mastodon
• Reddit
• LinkedIn

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

[greptile-apps]

Confidence Score: 5/5

The code change is a one-line allowlist addition with full test coverage; safe to merge once the changelog is trimmed to only the GLM-5.2 entry.

The functional change is minimal and correct — one new prefix check in a pure transformation function with no side effects, matching the established pattern for DeepSeek V4. Tests cover both the bare and provider-prefixed model name and include proper provider registration setup. The only issue is in the changelog, which contains extra entries from other contributors that appear unintentional.

core/changelog.md — contains 8 extra entries attributed to other contributors that should be removed before merging.

Important Files Changed

Filename	Overview
core/providers/openai/utils.go	Adds glm-5.2 prefix check to supportsMaxReasoningEffort; minimal, correct, and consistent with the existing deepseek-v4 pattern.
core/providers/openai/chat_test.go	Adds GLM-5.2 test cases (bare and provider-prefixed) to both reasoning-effort normalization test functions; provider registration/cleanup via defer is correct.
core/providers/openai/responses_test.go	Adds equivalent GLM-5.2 test cases to the responses API reasoning-effort normalization test; structurally correct.
core/changelog.md	First added line is correct for this PR; the remaining 8 added lines are attributed to other contributors and appear to be unintentionally pulled in from a local branch sync.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["normalizeOpenAIReasoningEffort(model, effort)"] --> B{effort == 'max'?}
    B -- No --> C[handle other cases]
    B -- Yes --> D{supportsMaxReasoningEffort?}
    D -- "Yes: deepseek-v4* OR glm-5.2*" --> E["return 'max' ✅"]
    D -- No --> F{supportsOpenAIXHighReasoningEffort?}
    F -- "Yes: gpt-5.2*, gpt-5.4*, etc." --> G["return 'xhigh'"]
    F -- No --> H["return 'high'"]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A["normalizeOpenAIReasoningEffort(model, effort)"] --> B{effort == 'max'?}
    B -- No --> C[handle other cases]
    B -- Yes --> D{supportsMaxReasoningEffort?}
    D -- "Yes: deepseek-v4* OR glm-5.2*" --> E["return 'max' ✅"]
    D -- No --> F{supportsOpenAIXHighReasoningEffort?}
    F -- "Yes: gpt-5.2*, gpt-5.4*, etc." --> G["return 'xhigh'"]
    F -- No --> H["return 'high'"]

Loading

_{Reviews (3): Last reviewed commit: "Merge branch 'dev' into fix/glm-5.2-max-..." | Re-trigger Greptile}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@core/changelog.md`:
- Around line 8-9: Remove lines 8-9 from the changelog.md file as these entries
document fixes for ChatAssistantMessageToolCall.extra_content preservation and
Anthropic prompt-cache diagnostics, which are out of scope for this PR focused
on GLM-5.2 support for max reasoning effort. If these changes are part of other
features, they should be committed to their respective feature branches with
their own changelog entries. Additionally, line 8 uses excessive exclamation
marks (3 total) which deviates from standard changelog style — ensure any
retained changelog entries follow consistent punctuation conventions.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 401e1cc1-3850-4589-9f21-ef92be3924ab

📥 Commits

Reviewing files that changed from the base of the PR and between d9c3a51 and 526254e.

📒 Files selected for processing (1)

• core/changelog.md