Agent 4 has been running noticeably slower for me, and the results often don’t match the prompt closely. I’ve had to repeat or rewrite prompts multiple times to get usable output.
Is there a known issue with Agent 4 right now, or are there any planned updates to improve speed and prompt accuracy?
I’d be happy to share examples if that would help.
There have been widely reported issues with Anthropic/Claude/Opus. This is beyond Replit control, but as these models are heavily used by Agent, it is affecting Replit at times.
This is true - Claude is having issues with not having enough compute for all the new users, thats why they have been giving longer and longer timeouts to subscription users, I canceled my subscription last week and just switched to Codex, GPT 5.5 seems better than Claude anyway. I’m sure that this will affect anyone who is using Claude API for their wrappers.