Google Models Often Broken in Chat/Composer Output

Flash 2.0


Flash 2.0 thinking

So I am using a Google key to access their models. I generally have to stick to the flash model because Google Pro Exp and thinking usually never output their answers correctly. Usually the code blocks are messed up, and generally the output in composer doesn’t work well.

So I stick to flash but sometimes it doesn’t work, failing to show the inserts correctly (1st image)

It would be greatly appreciate to expand composer/agent feature to work well with different model providers, like Google. Especially since google has a super generous free quota

1 Like

I’m pretty sure they’re working on this presently. Experimental built-in support for the Gemini models was added just before the seasonal break, but it’s definitely pre-alpha for the moment while they work on them:

The models do feel rather high-temperature from experiences so far, possibly rather harder to tame than Claude or 4o, but no doubt the model-wrangling will improve as the team gets to grips with the models’ idiosyncracies.