Can't Use BYOK GLM 5.2 from Scaleway in Cursor

Support Bug Reports
1

I can’t use BYOK GLM 5.2 from Scaleway in Cursor

When I add “glm 5.2” in Cursor (name has to be exact as Scaleway requires it):

Scenario 1: Use exact name glm-5.2

  • Cursor detects it as “built-in model”
  • Auto-adds suffix: glm-5.2 → glm-5.2-high
  • Scaleway rejects: “MODEL NOT FOUND”

Scenario 2: Use custom name glm-5.2-scw

  • Cursor sends exact name (good!)
  • Scaleway rejects: “MODEL NOT FOUND”

Result: GLM-5.2 is completely unusable in Cursor IDE.

Why This Is Cursor-Specific

Setup Works?
Scaleway Console (direct API) :white_check_mark: Yes
curl command :white_check_mark: Yes
Custom proxy (translates names) :white_check_mark: Yes
Cursor :cross_mark: No

This proves Scaleway API works correctly - Cursor’s model name modification is the issue.

What I’ve Tried

  • :cross_mark: Database editing - Cursor overrides at runtime
  • :cross_mark: Custom model names - Scaleway doesn’t recognize them
  • :cross_mark: Override OpenAI Base URL - Still modifies name
  • :cross_mark: Built-in GLM 5.2 model - Still adds -high suffix

Requested Fix: Disable Model Name Modification for Custom Providers

When “Override OpenAI Base URL” is set, send exact model names without modification.

Reproduction Steps

  1. Settings → Models → Enable “Override OpenAI Base URL”
  2. URL: [Scaleway base URL]
  3. API Key: [Scaleway API key]
  4. Add custom model: glm-5.2
  5. Send “hi” in chat
  6. Error: model 'glm-5.2-high' not found
4

Hey @n11o it is a known but right now and the team is already working on it. It comes down to the fact that since GLM is an available model inside cursor roadster, Cursor intercept the BYOK request and pass it as glm-5.2-high, that is bit supported by some BYOK provider, fix is coming

6

The exact-name issue is the important part here.

For BYOK providers, Cursor probably needs a stricter “pass through this model id exactly” path. Once a hosted provider has its own naming rules, even a helpful suffix like -high can turn a valid model into a completely different, invalid request.

7

thanks for the info :folded_hands: . Pls keep us posted

9

Hey, thanks for the detailed report with steps. That really helped.

Confirmed, this is a bug on our side, not an issue with your key or Scaleway config. When a custom model is named glm-5.2, Cursor resolves it as a built-in model and appends the reasoning suffix, so it sends glm-5.2-high to the provider. Scaleway doesn’t recognize that name, so you get MODEL NOT FOUND. With a custom alias, there’s another issue. Cursor sends the name as-is, but Scaleway only accepts glm-5.2. So yeah, there isn’t a clean client-side workaround right now.

We’ve reported the issue and we’re tracking it. I can’t share an ETA for the fix yet. As a temporary option, you can use a different model from the same provider until the fix ships.

I’ll post an update in the thread as soon as I have one.