The 2m context window is great from the latest Gemini models, but 8k output tokens is far less than competitors models. Of course for text data, you can ask the model to continue, however for json validated outputs, it’s not as straightforward.
1 Like
Hi,
Welcome to the forum.
Do you have a good use or example? Eventually, it might be better to open an issue here: https://issuetracker.google.com
Cheers.
I’m running into the same issue, I want to use Google Gemini 2.0 Flash to take call recordings from my call center and transcribe them and provide classification and or feedback/context and return a structured json with the desired format, currently for a 50 minute conversation the 8K context is exausted at minute 15, althouhg on Google AI Chat I can tell it to. “Continue” and it will keep writing the json, its not so feasible since I would have to make sure to concatenate the json and validate it, not sure if there is a better way to achieve this.