Comment by fidotron
6 hours ago
The Claude models definitely felt more susceptible to moods, like you could leave them for a few hours, come back and it suddenly was unable to do things which it was doing just earlier, which tellingly is never an experience I've had with an open model.
Honestly I lost patience with Anthropic both clearly messing around with things like this and their agitation over regulation. They aren't good actors, and quite why so many blindly trust them with their company crown jewels is a mystery.
Claude Code's prompt cache expires after 1 hour.
The cache shouldn't affect inference. It is purely an I/O optimization.
I think it should, as you dont need to use the encoder layer on the new tokens, you just read the embedding from the cache. that's why cache reads are cheaper
1 reply →
Is that from start of a new conversation per conversation?
It's supposed to be for token optimization (https://code.claude.com/docs/en/prompt-caching), but are people experiencing degraded performance when you let Claude Code sit for hours/days and come back?
1 reply →