This bug is in the Cursor Agent (chat/agent mode), not the IDE itself.
When a user specifies a model name that is approximate or uses a slightly different format (e.g. “grok-4.5” or “grok-4/5” instead of the exact internal identifier), the agent returns a hard error with zero actionable guidance:
“AI Model Not Found: Model name is not valid: ‘grok-4.5’”
The problem is two-fold:
The agent requires the exact model string but never tells users what the valid strings are.
There is no fuzzy-matching or closest-match suggestion — it just hard-fails.
This is especially frustrating because Cursor itself announced “Grok 4.5 is now available” on the forum, yet there is no way for the user to know what exact string to use when specifying it as a model in the agent config or chat.
Steps to Reproduce
Open Cursor IDE and start the Agent (chat mode).
In the model selector or agent config, type or specify “grok-4.5” (or “grok-4/5”) as the model.
Send any message/prompt to the agent.
Observe the error: “AI Model Not Found: Model name is not valid: ‘grok-4.5’”
Note that the error provides no list of valid model identifiers and no suggestion of what to use instead.
Example error Request ID: serverGenReqId_0b1a816a-4b49-4879-9be9-3926432a4219
Expected Behavior
Option A (ideal): The agent fuzzy-matches the input to the closest valid model identifier and proceeds, with an optional informational notice such as: “Using ‘grok-4.5-default’ as the closest match for ‘grok-4.5’”.
Option B (minimum acceptable): The error message displays a list of all currently valid/supported model identifiers so the user can correct the input and retry.
Currently the error is a dead end with no recovery path.
Operating System
MacOS
Version Information
Cursor IDE - latest version (July 2026)
For AI issues: which model did you use?
grok-4.5 (attempted; exact valid identifier unknown to user)
For AI issues: add Request ID with privacy disabled
Hey, thanks for the detailed report. One important detail: in the model selector you can’t type an arbitrary string. Models that Cursor provides must be picked from the dropdown list. Manual model IDs only work for custom models via BYOK with your own API key. That’s why manually entered strings like “grok-4.5” or “grok-4/5” fail with this error.
The right way is to open the model dropdown in the chat and select “Grok 4.5” from the list, don’t type the name by hand.
One more thing: based on what we see, the email used on the forum is on the Free plan, and that plan has a limited set of models. It’s possible Grok 4.5 simply won’t show up in the dropdown. If the model isn’t in the dropdown at all, it’s more likely a plan access issue, not a naming format issue.
On your suggestions about fuzzy match and showing the list of valid models in the error text, I agree the current error isn’t very helpful. I’ll pass this on as a UX improvement request. I can’t promise an ETA.
Thanks for clarifying, that changes things. When you run it from Linear, it uses Cloud Agent, and the model works differently there than in the IDE. You can’t set it by typing a string like “grok-4.5” in the task text. Cloud Agent takes the model from the Default model setting in the Cloud Agents dashboard: Cursor dashboard > Cloud Agents > Settings. Pick “Grok 4.5” from the list there, and agents started from Linear will use it. One important detail: the Cloud Agents default model must support Max Mode.
One more thing: sometimes agents started from Linear don’t pick up the Default model from the dashboard. In that case, you can set the model directly in the task or in a comment when calling @Cursor using the inline syntax [model=...], which overrides the dashboard default. Just don’t use an approximate string. Plain “grok-4.5” can trigger the same “not valid” error. It’s best to copy the exact model name from the dropdown in the dashboard and use that.
If Grok 4.5 doesn’t show up in the Default model list at all, it’s most likely an access or plan issue, not a naming format issue. Cloud Agents from Linear are usually set up by a team admin, so double check the run is using the correct work account on the right plan, not a personal Free account.
On your suggestion about fuzzy matching and listing valid models in the error text, I agree the message isn’t very helpful right now, and I passed it along as a UX request. I can’t share an ETA. Let me know how it goes.
Yes, the UX improvement would be very helpful. For example, with Grok, so many model names, and each platform exposes them differently. Where can I see an exact list of the Grok models supported by Cursor?
In the IDE: the model dropdown in chat. This is the most reliable source for the IDE because it shows what’s actually available on your plan. You can copy the names from there as-is.
For Cloud Agents, including runs from Linear: the Default model dropdown in Cursor dashboard > Cloud Agents > Settings. Keep in mind the model lists for the IDE and Cloud Agents are intentionally different.
A couple notes if you’re using the API or SDK:
Cursor.models.list() in the SDK and GET /v1/models return the exact model ids, which is handy for programmatic selection: Available models | Cursor Docs.
/v0/models in the Cloud Agents API only returns a limited list and doesn’t always show every working model, so don’t rely on it as a complete reference. For Cloud Agents, use the dropdown in the dashboard.
Tip: don’t type model names by hand. Copy the exact id or name from the dropdown to avoid the not valid error.