GLM-5.3-Flash degraded: slow or stalled responses
Resolved
Nov 20, 2026 at 1:59am UTC
The fix is deployed to all GLM-5.3-Flash capacity and responses are back to normal. We will keep monitoring closely.
Affected services
Updated
Nov 20, 2026 at 1:39am UTC
The fix is live on part of GLM-5.3-Flash capacity and we are completing the rollout now. Responses should be returning to normal; we will update when it is fully complete.
Affected services
Updated
Nov 20, 2026 at 1:13am UTC
We identified the cause: certain very long tool-heavy conversations take minutes to prepare, which holds up other requests on the same server. A fix is tested and being rolled out now. You may still see occasional slow responses on GLM-5.3-Flash until it completes.
Affected services
Created
Nov 20, 2026 at 12:58am UTC
Since about 23:55 UTC some GLM-5.3-Flash requests have been slow to start or have stalled. We have identified the cause and are restoring capacity now. Other models are not affected.
Affected services