fix(zai): resolve model list for China API users (endpoint-aware)

[Icather]

What does this PR do?

Fixes Z.AI model listing for two bugs:

Bug 1 -- provider_model_ids resolves the correct endpoint URL (e.g. China's open.bigmodel.cn) via resolve_api_key_provider_credentials, but fetch_models still hits the hardcoded international URL (api.z.ai). China API keys are rejected, falling back to fallback_models (only 2 models).

Bug 2 -- Z.AI's /v1/models omits universally-free Flash models (glm-4v-flash, glm-4.6v-flash, glm-4.1v-thinking-flash, glm-4.5-flash, glm-4-flash-250414) even though they accept real API calls. The fix appends them via an override on the Z.AI ProviderProfile.fetch_models.

Related Issue

• Closes fix(models): merge Z.AI curated models with live API results #46623 (earlier approach merged full curated list -- this one is endpoint-aware and only appends verified free models)

Type of Change

• 🐛 Bug fix

Changes Made

• plugins/model-providers/zai/__init__.py -- Added ZAI_FREE_MODELS constant (5 universally-free models). Overrode fetch_models to append these after the live API response with dedup.
• hermes_cli/models.py -- Added base_url swap in the generic live-fetch path so fetch_models hits the resolved endpoint instead of the hardcoded profile default.

How to Test

1. Configure a Z.AI API key via the China platform (open.bigmodel.cn)
2. Open Dashboard → Models page → select Z.AI provider
3. Verify 12 models appear (7 live + 5 free Flash)
4. Select a free vision model (e.g. glm-4v-flash) as auxiliary vision model -- confirm it works

Checklist

Code

• Commit messages follow Conventional Commits
• Searched for existing PRs (closed fix(models): merge Z.AI curated models with live API results #46623 is superseded by this)
• PR contains only related changes
• Tested on: Windows 11
• Run pytest tests/ -q -- N/A (no new tests; fix is integration-level)
• Added tests -- N/A (requires live Z.AI API key)

Documentation & Housekeeping

• Updated relevant documentation -- N/A
• Updated cli-config.yaml.example -- N/A
• Updated CONTRIBUTING.md or AGENTS.md -- N/A
• Considered cross-platform impact -- Yes (fix applies to both international and China endpoints)
• Updated tool descriptions/schemas -- N/A

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds support for Z.AI “universally-free” models that are not returned by the provider’s /v1/models endpoint, and updates CLI model fetching to respect a resolved per-region base URL.

Changes:

• Add a Z.AI free-model allowlist and a wrapper around fetch_models() to append them.
• Remove glm-5.2 from Z.AI fallback models.
• In hermes_cli, temporarily apply the resolved base_url while fetching live models.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
plugins/model-providers/zai/init.py	Adds free-model list and wraps fetch_models to append those models.
hermes_cli/models.py	Ensures fetch_models uses the resolved (possibly regional) base_url.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[teknium1]

Thanks for the targeted Z.AI investigation. The resolved-endpoint part has since landed on current main in 1b962f001: hermes_cli/models.py:2485 passes the credential-resolved base_url, and providers/base.py:189-194 uses it for /models.

Problems

• The proposed wrapper at plugins/model-providers/zai/__init__.py:14 does not accept base_url, but current main now calls fetch_models(api_key=..., base_url=...) at hermes_cli/models.py:2485; salvaging this unchanged would raise TypeError.
• Its call at plugins/model-providers/zai/__init__.py:16 also omits base_url, which would discard the resolved China endpoint.
• Current main already has ZaiProfile (plugins/model-providers/zai/__init__.py:85), so an override there is preferable to replacing a method on one instance.

Suggested changes

• Rework the remaining free-model addition as a ZaiProfile.fetch_models override that accepts and forwards base_url, normalizes an empty live result, and deduplicates the appended IDs.
• Add an offline regression test for the resolved URL and empty-catalog path.

Automated hermes-sweeper review.

[teknium1]

Current main invokes API-key profiles as fetch_models(api_key=..., base_url=...) in hermes_cli/models.py:2485. This keyword-only wrapper does not accept base_url, so salvaging it unchanged raises TypeError; add and forward that parameter.

[teknium1]

Forward the resolved base_url here. Current ProviderProfile.fetch_models uses that override to select the endpoint (providers/base.py:189-194); omitting it would fall back to the profile's international default and undo the China routing fix.