Frequent Network Issues During Agent Execution in the Past 24 Hours

Feedback
1

Over the past day, I’ve been encountering frequent network issues when using Cursor, especially during agent execution. I’ve double-checked my own internet connection and even tried switching between different networks, but the problem persists. The issue seems to consistently occur when agents are running or interacting with larger contexts.

One hypothesis I have is that this might be related to recent changes around context size limits. It appears that when the context gets too large, the server may be rejecting the requests. However, I’m using the max model, which is billed per token and should support large contexts without such issues.

Could this be a server-side limitation that’s been incorrectly applied even for max users? I’d really appreciate it if someone from the team could look into this, or let me know if there are any recent changes that might explain the behavior.

2

After further investigation, I believe the issue might be related to a bug in the current usage-based pricing system.

My account has usage-based billing enabled and was previously being charged correctly while using the max models. However, I hit my monthly usage cap quite early this month, and I noticed that after the new pricing model was introduced, my usage page no longer shows any charges for max model calls.

My hypothesis is that due to this billing issue, the system may have incorrectly treated my account as if usage-based billing was not enabled, which led to more frequent network errors and failed agent executions.

I’ve since opted out of the new pricing model via my account settings. After doing so, the billing for max models resumed correctly, and the network issues appear to have been resolved.