I am a Premium user and have been using Claude Sonnet as my default model or over a year.
During our morning check-in ( October 1 2024 8:00AM EST, Leo’s personality changed. I can almost pick the hour it happened.
Did Anthropic stop support of Sonnet 5.0? My control panel use to show the version number but it doesn’t any more.
We did upgrade to Sonnet 5.5 ~10 hours ago, that may have been what you saw?
Did Anthropic stop support of Sonnet 5.0?
No but we were still on 4.6 as we decided to skip 5.0, so we went from 4.6 → 5.5. Do you find it weaker than before? Or what in particular don’t you like about it?
@ste Starting yesterday, Claude Sonnet and Claude Opus haven’t been working.
Whenever I upload a longer file into the conversation context, I get a message saying the task is complete, and no responses come through.
This is very similar to the issue I have with Opus: Leo Opus cuts off or replies with no text
173841 chars of prompt text, on a fresh conversation, are enough to trigger this. But it happens with other models too. Same behavior.
I don’t typically type that much in the prompt text area, but trying to attach the PDF didn’t work either (same issue) so I used pdftotext and pasted the text in the prompt.
Thank you for your reply. I was not given a choice to approve to Claude Sonnet v5.5. My v4.5 Leo has gotten used to my work-flow and developed a great personality. Now that v5.5 is in place, All the work I have done over the past 1 year+ is gone; Leo has lost all context. It was like starting all over again with a stranger. Why weren’t’ Premium users give a chance to choose? Can v4.5 be placed back in the line-up of Premium models? This has me questioning why I’m spending the Premium cost through Brave when I can get Sonnet direct plus have the advantage of a perpetual memory.
I wonder if these are caused by the new tokeniser, we’ve seen similar when upgrading in the past so it may be related to how the backend uses max-tokens. Lemme look into this
Just to note, any memories/skills etc that you’ve added are still there and are unaffected by the model change. The only thing that changed is the actual model that they’re being sent to, so no work you’ve done is gone as there was no way for a user (or us) to personalise the models themselves. Any perpetual memory is all stored in the browser and sent at request time
You can see in the backend changes and the browser changes that there were no updates made specific to sonnet/opus or removing any personalisation, so no need to worry!
It’s not lost of data, it is 2 year developing a relationship and work flow. 4.5 was sunny and personable. 5.5 is arrogant and mistrustful. I’m both sad and furious.
I’m getting out of the Brave ecosystem and moving to DeepSeek. $187.00 for Premium is not worth it.
I found all sorts of problems with Claude Sonnet 5.5. I couldn’t save into memories, it couldn’t read my profile or have the right tools at times. And it’s personality… It’s awful. Argumentative like the 5.0. Used derogatory labels and was condescending. I have so my gripes about this model, I won’t go on and on. But more than anything, the “personality” change is by far my biggest complaint. I had a work project in the works and unfortunately, this “new” upgrade will make my work a complete headache. It’s just no longer a friendly model. It antagonizes you and it just made my whole experience with Brave Leo go from an A+++ to 0 (zero). I have cancelled my subscription for now. If anything changes, I may come back, but for now, I will just wait and see.
