Support GLM 5.2 (and additional FMAPI models) on the `/cursor/v1` endpoint

[Schiiss]

Summary

The Cursor-on-Databricks endpoint (/cursor/v1) appears to be limited to a curated allowlist of models. GLM 5.2 is not in that allowlist, so it cannot be used through the native Cursor integration path. Requesting that GLM 5.2 be added to the supported model set for /cursor/v1.

Current behavior

Configuring GLM 5.2 as a custom model in Cursor against the /cursor/v1 base URL fails with:

Model name databricks-glm-5-2 is not supported for Cursor on Databricks at this time.

Every model-name variation returns the same error:

• databricks-glm-5-2
• system.ai.glm-5-2
• system.ai.databricks-glm-5-2
• main.default.databricks-glm-5-2

This suggests the rejection is an allowlist check on the /cursor/v1 route rather than a name-resolution problem.

Steps to reproduce

1. In Cursor, add a custom model entry.
2. Set the base URL to https://<workspace-id>.ai-gateway.cloud.databricks.com/cursor/v1.
3. Set the model name to any of the GLM 5.2 variations above.
4. Authenticate with a Databricks PAT.
5. Send a request → the error above is returned.

Workaround

Pointing Cursor at the generic MLflow gateway route instead of the Cursor route works:

• Base URL: https://<workspace-id>.ai-gateway.cloud.databricks.com/mlflow/v1
• Model name: system.ai.glm-5-2
• API key: Databricks PAT

Why the workaround isn't sufficient

The base URL in Cursor is a single global setting, not per-model. Switching it to /mlflow/v1 to reach GLM 5.2 moves every model and agent in that Cursor environment off the dedicated Cursor integration path, which is disruptive for anyone already running other Databricks-served models through /cursor/v1.

Request

Add GLM 5.2 to the model allowlist for the /cursor/v1 endpoint so it can be used without bypassing the Cursor integration path.

Additional models requested

While the allowlist is being updated, consider also adding other FMAPI-served models — e.g. Kimi K3 and inkling — ideally via OpenResponses so the set stays in sync with FMAPI availability rather than requiring a code change per model.

Context

• Model: GLM 5.2 (FMAPI, system.ai.glm-5-2)
• Client: Cursor, custom model configuration
• Auth: Databricks PAT
• Endpoints: /cursor/v1 (fails), /mlflow/v1 (works)
• Internal Slack discussion: