I've been using Gemini Advanced 2.5 Pro and noticed that during long conversations, it seems to forget a lot of the earlier information I fed it. I spent two hours prepping, inputting loads of info and rules, but it's like it only remembers the more recent stuff. I thought 2.5 Pro had a huge context window—but something seems off now. Has it been reduced? What's going on? I rely on it for content research and writing and I'm starting to feel frustrated because I pay for multiple AI models but they all seem to be losing their effectiveness. Are they just shifting focus to other uses now?
4 Answers
Why is this even being discussed here? This really belongs in the Gemini forum, not OpenAI.
How are you feeding it URLs anyway? As far as I know, Gemini 2.5 Pro can’t access web content beyond Google search results.
I’ve been loading it with a ton of specific info, which takes quite a while to prepare. Had a method that worked well, but it seems like it’s just dropping earlier parts of the conversation.
Actually, it still retains the same context window, but benchmarks suggest it's not performing as well as before—up to 30% worse for remembering long conversations compared to its previous version.
So just to clarify, it can still take in long prompts, but it struggles to keep track of everything over time? That’s frustrating!
Sounds like you're getting into some serious detail. What sort of things are you putting in there?