HACKER Q&A
📣 rvnx

Did Claude Sonnet 3.5 suddenly became worse for you today?


Did Claude become worse today ?

I have been using it as a programming assistant. It feels like it became really bad today ?

Lazy, not writing logical code, just non-sense useless code and the style of it is reminiscent of GPT-4o for some reason.

It is so strange, it suddenly came during the night, like if there was a system update.

Is it just me or you are affected too ?

This is example of code it outputs:

  if (decision === '') {
    if (code === '') {
      break;
    }
    code = code.slice(0, -2);
    if (code === '') {
      code = '';
    }
    continue;
  }

  👤 phyalow Accepted Answer ✓
It's been a popular topic for the last couple of weeks on /r/claudeAI/.

I work via the API (console.anthropic.com/workbench) which is still showing a static model number and has less safety filtering applied to it than interacting with the front end. Performance seems mostly acceptable still for me.

Its obvious they have had trouble meeting compute demand, so it wouldnt suprise me if they have quantized the model or something to help them out on capacity issues for "Pro" users interacting via the front end.

👤 readyplayernull
Also recently, Poe's Assistant, which is free, used to reply with direct to the point answers which I loved, but just a few weeks ago it became as chatty as ChatGPT. It's annoying having to read all the extra wording. Once again, don't fix what ain't broken.
👤 sys32768
I noticed its AI IQ dropping this past week. I use a paid subscription via POE.COM.
👤 jaggs
Could it have anything to do with this? https://abcnews.go.com/US/wireStory/authors-sue-claude-ai-ch...
👤 ldjkfkdsjnv
My guess is they downsized the model
👤 palashkulsh
In my personal experience, it happened before as well but got better in few days
👤 ftyhbhyjnjk
I have read many tweets since yesterday saying they think sonnet 3.5 have been nerfed.