Ever feel like you’re talking to Dory from Finding Nemo when having a deep conversation with ChatGPT?
You have a long, productive chat, laying out all the details, and then… BAM. It misses details from 10 messages ago or starts making stuff up (hallucinating).
Frustrating, right?
A big reason this happens is the AI’s “context window”. Basically, its short-term memory.
Think of it like this:
Every time you send a message and the AI replies, it re-processes the entire prior chat history (up to its limit).
Every word in that back-and-forth consumes tokens, basically eating up AI’s brain space (the context window).
So even ChatGPT Plus’s 32,000 tokens can fill up faster than you think.
This has been a pain point for me. After a few chats, the AI forgets the requirements at the beginning
Recently I’ve been digging into Google Gemini. It’s become a favourite. It def has quirks which I’ll talk about below. But it does tackle this memory issue head-on.
The Standout Feature: Massive Context & Raw Power
Gemini 2.5 Pro offers a 1 MILLION token context window. That’s around 750,000 words.
Before you tell me that your fav AI tool, like Merlin, supports Gemini 2.5 Pro, let me ask you something…
Are you sure they didn’t set a cap on the context window?
In fact I’ve used Merlin a lot, and I do feel they set some kind of limit. Likely close to ChatGPT’s (32k). Understandable since these AI ain’t cheap.
The point is simply using Gemini 2.5 Pro in your fav AI wrapper may not solve the problem. Especially if THEY are the one paying for usage. I won’t be surprised if they set a backend context limit.
Now, let’s be clear: for most everyday chats that don’t stretch on forever, this isn’t a problem.
Remembering the last 5-10 messages is often plenty for good results. We’re talking about the long, complex conversations where deep memory becomes critical.
Anw, what does it mean to have a huge context window?
You can have incredibly long and detailed conversations. Like:
- brainstorming complex strategies
- providing detailed client docs to brainstorm
- getting detailed marketing/business advice
- drafting long-form content with lots of iteration
And it remembers.
It doesn’t use RAG (indexing and retrieving chunks).
It processes the entire chat with raw power, giving it near-perfect recall. And so reduces hallucination.
Case in point: This very post draft!
It involved lots of back-and-forth brainstorming and iteration with the AI. The total chat is around 13,000 words and 85,000 characters. That’s way beyond the memory limits of tools like ChatGPT.
Another Big Plus: Unrestricted Custom Instructions
You know that “Custom Instructions” or “System Prompt” section where you tell the AI background info?
Most tools limit this severely:
- ChatGPT: 8,000 characters
- Perplexity: 4,000 characters
- Taskade: 10,000 characters
- Merlin: ~4,000 characters
- Straico: No hard limit, but you pay token costs for the instructions on every input.
I think Gemini is unlimited. I haven’t hit a hard limit yet.
I’ve pasted thousands of words of client background or project specs into the instructions. It takes it like a champ.
Even if there is a limit, the 1M window means you can just paste long instructions at the start of a conversation.
Rapid Feature Rollout
Google moves fast.
Just this week while I was prepping my thoughts for this post, they rolled out video gen and some AI audio thing. Feels like it’s evolving quickly.
They’ve got money. So they can offer things like unlimited AI and huge context window.
BUT… It’s Not All Sunshine and Rainbows (The Quirks)
Okay, here’s where Gemini feels a bit… unfinished:
No Searchable History: This is baffling. You can’t search your past conversations. Seriously, Google? No search?
No Folders: Forget organizing your chats into folders. Doesn’t exist.
Workaround: You can pin important chats and rename them. Then you can scroll your pins or use search (Ctrl+F). It’s clunky.
Tip: use keywords in the title to help you search faster. E.g: + +
No Desktop App: It’s web-only. I use Wavebox to create a “desktop app” so I can open it with keyboard shortcut. You can also use Edge for this.
No Keyboard Shortcuts: Basic stuff like Cmd+N for a new chat or a shortcut to clear context? Nope. You’re clicking around.
So, Why do I use it?
Despite the UI/UX annoyances, the core strength – that massive context window + the quality of the Gemini 2.5 Pro model – keeps me coming back.
When I foresee a long chat or need to have a huge input, Gemini Advanced is my go-to.
It’s not replacing some other wrappers like Straico or Merlin (which is fantastic value for other models). But it’s earned its place for specific, demanding jobs.
Your turn!
- Have you tried Gemini (Free or Advanced)?
- Hit similar context walls with other AI?
- Found other tools that handle long conversations well?
