Response Cutoff/unfulfilled message

Sometimes, when prompted for reasoning models like Claude sonet 4.6, due to network issue or focus change, streaming response is raised. After refreshing the page, some of the response/reasoning is displayed, but the rest of it is not. Sometimes the whole message is empty, sometimes it shows empty message but shows reasoning. We can re-run the prompt, but the tokens are all ready deducted for the previous request, and re-ruining the request will drain more tokens. For costly models like sonet 4.6, it could cost approximately the whole 4 hour limit.

I think after firing the prompt, the response, wether streaming to client (active) or not (broken due to network issue, inactivity), should always get back the model response after reload/refresh. I have observed it 5-6 times as I prompt the AI and go to do other things.

Attaching a latest screenshot of this happening to me in between streaming response.