What's going on with Gemini?

Google is in a strange spot right now. They've got arguably the deepest research bench in the industry, their own custom silicon, and effectively unlimited money - and yet most developers I talk to barely touch Gemini day-to-day. The recent Google I/O announcements crystallised a lot of what I find confusing about their AI strategy, so I wanted to write down where I think they actually stand. The state of play The consensus seems to be that currently Anthropic and OpenAI are very much in the lead for frontier model intelligence, with each of those two labs trading blows every month. This may change in the near future - if Anthropic releases Mythos-class models that OpenAI doesn't have an answer to - but right now I think most practitioners would agree that GPT5.5 and Opus 4.8 are roughly in the same ballpark. After that, you have Google, with Gemini 3.1 Pro being in benchmarks ahead of the Chinese models but behind the flagship Anthropic/OpenAI models. In my personal experience though…