Cursor Router classification error falsely triggers ConnectError: [resource_exhausted] and fails to fall back (Ticket T-E74983)

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

We identified a critical bug in Cursor Router’s fallback logic. When an unrecognized model classification event occurs, the Router treats different accounts under the same Enterprise organization inconsistently—allowing some users to fall back gracefully to default models while throwing a hard ConnectError: [resource_exhausted] error for others, completely blocking developer workflows on our Enterprise account.

Support ticket T-E74983 confirmed via backend logs that this is NOT a quota exhaustion issue, but a false resource_exhausted error surfaced by the Router when its classifier fails hard instead of falling back.

Steps to Reproduce

On an Enterprise request-based plan, select Auto mode in Composer or Chat.

Ensure the user’s monthly premium request quota is exhausted or hitting a boundary.

Submit a prompt or code generation request.

Result: Instead of gracefully falling back to a default/cheaper model or handling the classifier error gracefully, the request fails immediately at the transport layer with:
[composer] Error in AI response ActionRequiredError: You’ve reached your monthly usage limit / ConnectError: [resource_exhausted].

Expected Behavior

When the Router encounters an unrecognized model classification event or quota limit under Auto mode, it should fall back gracefully to default/lower-cost models (as it did for other team members in our org) rather than throwing a hard ConnectError: [resource_exhausted] and blocking the request entirely.

Screenshots / Screen Recordings

Snipaste_2026-08-12_10-15-27.jpg
Snipaste_2026-08-12_10-15-27.jpg894×548 21.9 KB

Operating System

MacOS

Version Information

Version: 3.15.6
VS Code Extension API: 1.128.0
Commit: a1f686545fd0ce8917bbd2449f733551a9bce420
Date: 2026-08-06T01:41:03.876Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.291
OS: Darwin arm64 22.6.0

For AI issues: which model did you use?

Auto (Cursor Router)

For AI issues: add Request ID with privacy disabled

Blocked Request (Failed with resource_exhausted): 917a4f9b-cbc7-482a-80c3-4578ee755c79

Working Request (Same org/status, fell back gracefully): fc0ce7f9-9aa0-456a-b711-37e8773cce4c

Does this stop you from using Cursor

Yes - Cursor is unusable

6

Hey there!
This isn’t a fault in Cursor Router. What you’re hitting is the expected behavior when an account has used up its monthly usage allocation for the cycle. ConnectError: [resource_exhausted] is just the transport-level form of the You've reached your monthly usage limit message.

Fastest unblock: your team admin can raise that member’s monthly limit in the team dashboard under Members, or it resets automatically at the next billing cycle. See Usage and limits and Members & roles.

Two quick notes: usage is tracked per member, so the two request IDs you shared are tied to different accounts, which is why one went through and one was blocked. And picking a specific model instead of Auto won’t get around it, since the limit applies to any model.

If a member is still blocked right after their limit is raised, send a fresh request ID from that attempt and I’ll take another look.