By skyforbes Hi everyone,
I've been using Gemini for a while, but recently, especially with the heavier models like Gemini 3, I've noticed a really frustrating issue regarding how it handles connectivity and active sessions.
The Problem:
When I give Gemini a complex prompt (especially on models that take a while to think/generate), I naturally want to switch contexts. For example, I might start the prompt on my PC and then grab my phone to check the result on the go, or simply lock my phone screen while waiting.
The Result:
The moment I disconnect, close the tab, or switch devices, Gemini aborts the generation and often deletes the entire chat.
It feels like the generation process is heavily dependent on the client-side connection. If the "heartbeat" with the browser/app stops, Google just kills the process immediately instead of letting it finish in the background.
The Comparison with ChatGPT:
This is a stark contrast to ChatGPT (especially with o1 or Deep Research modes).
* ChatGPT: I can start a task that takes 20 minutes, close the app, go make a coffee, switch devices, and the process continues running server-side. It doesn't care if I'm watching or not.
* Gemini: It basically demands that I stare at the screen. If I look away (disconnect), it punishes me by wiping the session.
Is this a technical limitation of how Google streams the tokens, or just a bad UX choice to save compute power? It makes using the "heavier/slower" models really impractical if I can't multitask.
Does anyone else experience this, or is it just me?
I've been using Gemini for a while, but recently, especially with the heavier models like Gemini 3, I've noticed a really frustrating issue regarding how it handles connectivity and active sessions.
The Problem:
When I give Gemini a complex prompt (especially on models that take a while to think/generate), I naturally want to switch contexts. For example, I might start the prompt on my PC and then grab my phone to check the result on the go, or simply lock my phone screen while waiting.
The Result:
The moment I disconnect, close the tab, or switch devices, Gemini aborts the generation and often deletes the entire chat.
It feels like the generation process is heavily dependent on the client-side connection. If the "heartbeat" with the browser/app stops, Google just kills the process immediately instead of letting it finish in the background.
The Comparison with ChatGPT:
This is a stark contrast to ChatGPT (especially with o1 or Deep Research modes).
* ChatGPT: I can start a task that takes 20 minutes, close the app, go make a coffee, switch devices, and the process continues running server-side. It doesn't care if I'm watching or not.
* Gemini: It basically demands that I stare at the screen. If I look away (disconnect), it punishes me by wiping the session.
Is this a technical limitation of how Google streams the tokens, or just a bad UX choice to save compute power? It makes using the "heavier/slower" models really impractical if I can't multitask.
Does anyone else experience this, or is it just me?