It's (probably) not the model.
It's this conversation.
Long chats wear out — the AI starts forgetting, repeating, drifting. Paste yours below and see the evidence in 10 seconds. If it's cooked, you'll get a fresh start that remembers what matters.
Even easier — paste a share link. On Claude or ChatGPT: open your chat → Share → Copy link, then paste the link here instead of the text. We pull the whole conversation server-side, nothing missed. (A share link makes that chat viewable by anyone with the URL — you can un-share it right after.)
🔒 We never store your conversation. Pasted text is analyzed right in your browser; a share link (and the written rescue) is processed on our server and discarded immediately after — never saved, never trained on.
How this level was decided
Built from your own conversation — the goal, the things that got lost, and where you left off. Edit anything, then paste it as the first message of a new chat.
“ChatGPT got dumber” — what's actually going on
When a conversation gets long, every AI assistant starts to wear out: it loses track of things you said early on, re-serves answers it already gave, and you catch yourself repeating instructions. Research calls this effect “context rot” — performance degrades as the conversation fills up, and it doesn't degrade gracefully. It's not your imagination, and it's usually not a model update either. It's this specific thread.
How to tell the difference
If a fresh chat handles your question well but this conversation keeps stumbling, the conversation is the problem. The check above gives you the evidence from your own chat: what you had to repeat, what it forgot, what it re-served. If the thread is healthy, it tells you that too — then the issue may genuinely be the task or the model, and starting over won't fix it.
The fix that actually works
Don't fight a cooked conversation — replace it. The handoff built above carries your goal, your constraints (especially the ones the AI kept dropping) and the current state of the work into a new chat. Two minutes, and you're working with a sharp assistant again.