Multiple AI models fail with “Fallback Model Disabled” in Cursor Pro

[FWKTechnologies]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I am experiencing an issue with multiple AI models in Cursor IDE.

All Claude models I have tested fail when I send a request. GPT-5.6 Sol also fails with the same issue.

Cursor displays the following error:

“Fallback Model Disabled
The suggested fallback model is disabled. Choose a different model to continue.”

The affected Claude models include:

• Claude Opus 5.5
• Claude Opus 5
• Claude Sonnet 5
• Claude Fable 5.1
• Claude Opus 4.8
• Claude Sonnet 4.6
• Other Claude models available in the model picker

I also tested GPT-5.6 Sol, and it fails as well.

The models are visible and selectable in the model picker, but requests fail when I actually try to use them.

I have not configured any personal provider API keys for Anthropic, OpenAI, Google, Azure OpenAI, or AWS Bedrock.

Please investigate the model access, model entitlements, fallback routing, and backend configuration for my account.

Steps to Reproduce

1. Open Cursor IDE.
2. Select a Claude model from the model picker.
3. Send a normal request.
4. The request fails and Cursor displays “Fallback Model Disabled”.
5. Repeat with other Claude models.
6. Select GPT-5.6 Sol.
7. Send a normal request.
8. The request fails with the same or similar error.

Expected Behavior

The selected supported model should process the request and return a response normally.

If the selected model is temporarily unavailable, Cursor should either use a working fallback model or provide a clear explanation of why the request cannot be processed.

Screenshots / Screen Recordings

Operating System

Windows 10/11

Version Information

3.18.9 (user setup)

For AI issues: which model did you use?

Multiple Claude models, including Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5, Claude Fable 5.1, Claude Opus 4.8, Claude Sonnet 4.6, and GPT-5.6 Sol.

Additional Information

The issue affects multiple model providers rather than a single model.

The models are visible in the model picker, but requests fail when they are selected.

No personal provider API keys are configured.

Please check whether there is an issue with model entitlements, model routing, fallback configuration, or backend access for my account.

Cursor profile name:
FWKTechnologies / FWK Technologies

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[Colin]

Hey @FWKTechnologies, thanks for the detailed report and screenshots!

This popup appears when your included usage for third-party models (Claude, GPT and similar, shown as “Other Models” on your usage page) has run out for the current billing cycle. When that happens, Cursor tries to switch the request to Grok 4.6, which uses your Cursor Models allowance instead. Since Grok 4.6 is turned off in your model settings, Cursor can’t make that switch, so every Claude and GPT request stops with “Fallback Model Disabled”. We know the message doesn’t make the real reason clear, and I’ve passed that feedback on.

You can check this at cursor.com/dashboard under Usage. “Other Models” should show as fully used.

To keep working, any of these will help:

1. Pick Auto or Composer from the model picker. These use your Cursor Models allowance and should work right away.
2. Or turn Grok 4.6 back on in Cursor Settings > Models, so the automatic switch can go through.
3. If you want to keep using Claude or GPT before your cycle resets, turn on on-demand usage in your dashboard spending settings.

[FWKTechnologies]

Hi Colin, thanks for explaining the fallback behavior.

However, there is still an important issue with the usage shown on my account.

I only used Claude 5.1 for approximately one request/task. The task was not even completed before I became unable to continue using Claude.

I have also never used GPT on this account, so I don’t understand why my entire “Other Models” allowance is already showing as 34.5M tokens / 100% used, and why GPT requests are also blocked.

Could you please check the usage attribution on my account and confirm exactly what generated the 34.5M tokens under claude-opus-5-thinking-high?

In particular:

• I did not intentionally use Claude Opus 5 Thinking High.
• I have never used GPT.
• I only used Claude 5.1 for a very limited amount of work.
• The task was not completed before the allowance was exhausted.

Could you please verify whether the usage was attributed to the correct model/account and whether there is any issue with the usage calculation or model routing?

Thank you.

[Colin]

Hey @FWKTechnologies, thanks for the screenshot. I was able to find your account and check the usage.

The 34.5M tokens all came from a single chat on Sep 22, started right after your billing cycle renewed. That chat was set to Claude Opus 5 (Thinking, High), and the agent ran about 225 steps over roughly 70 minutes before the Other Models allowance ran out. After that, the same chat continued on Auto.

The token count includes the full context the agent re-reads at each step, so one long agent task on a large model can use tens of millions of tokens even if you only sent a few messages. Nothing was switched to a different model behind the scenes. The usage page shows the model the chat was set to.

To keep working until your allowance resets on Oct 22:

1. Keep using Auto or Composer (your Cursor Models allowance is only about 8% used).
2. Or turn Grok 4.6 back on in Cursor Settings > Models, so Claude and GPT requests can switch automatically instead of stopping.
3. If you want to keep using Claude or GPT now, you can turn on on-demand usage in your dashboard.

For long tasks, a lighter model like Claude Sonnet 5 or Auto will stretch your included usage much further than Opus.