Mayfly models

Ear­lier this year I noted Google’s eager dep­re­ca­tion of Gemini models. Now there’s this, from my inbox:

Gemini 3.7 Flash is now dep­re­cated and no longer avail­able as an inde­pen­dent model in the Gemini API and Google AI Studio. All traffic is auto­mat­i­cally redi­rected to Gemini 3.8 Flash.

I assume the jus­ti­fi­ca­tion here is that 3.8 Flash is, as they say in trading card games, “a strict upgrade”—i.e. there’s no sit­u­a­tion in which 3.7 Flash is better than 3.8 Flash. And yet … how would they know?? The whole value of these models is that they are useful in a breath­taking variety of contexts … so, who’s to say there isn’t some niche appli­ca­tion that just broke, because 3.8 Flash has a subtly dif­ferent opinion of pic­tures of radishes than reli­able old radish-loving 3.7?

I will note that 3.7 was intro­duced in early August. It was barely two months old.

I don’t know … I wonder some­times if any of these com­pa­nies really expect any­body to build any­thing on top of these models at all. If all you’re doing is yakking with a coding agent: easy enough. But if you’re using them programmatically, as steps or gates in a larger process, their organic com­plexity makes the con­cept of “a breaking change” basi­cally impos­sible to reckon. Who knows? Who can know?? Nobody.

I suppose the really AGI-pilled response is, so what? If it breaks, get 3.8 Flash to fix it.

These are liquid platforms, for a liquid age.