They started early, publishing the paper that led to the LLM craze (https://research.google/pubs/attention-is-all-you-need/), and internally testing a model code named LaMDA (remember this: https://www.washingtonpost.com/technology/2022/06/11/google-...?) but by the time they released it as Gemini, it was significantly under performing competitors, and became a bit of a laghing stock. (https://www.bbc.com/news/articles/cd11gzejgz4o)

That was probably their low, and they've been catching up especially as the field has plateaued. They're making contracts to host LLM services, so they don't seem to be giving up, and the longer the plateau lasts, the less they'll be behind, unless they are slow to migrate to whatever the next step is.

They just conquered mobile hot off losing an antitrust case but getting away with it because AI exists, so they’ve never been in a stronger position. Both Google and Apple are fighting the EU to keep rival AI assistants at bay, they are on the brink of locking everyone else out almost globally.

I think giving away Gemma may be telling on the direction they wanna go long term.

Winning doesn't have to mean making the most money I hope. Maybe winning is paying homage to the years of organizing everyone's data and giving it back in the form of a highly optimized completely free capable model? Or am I too naive. Give a highly optimized code forge I can run on local affordable hardware and you will win in my eyes.

Giving away Gemma was the first time I thought google may have a heart after all. Or am I too naive.

They won a Nobel prize using it for something other than chatbots

No? The next Siri update will be powered by Gemini. Dozens of Google products integrate and deploy AI to billions of users.

I think they're concentrating on users and not benchmarks. Gemini is still the best model at plain old searches, like, for parsing images or "what's that poem with this and this in it". It seems to have much more breadth of reasoning than any of the other models, e.g. Gemini 3.1 is still the top model on simple-bench other than Fable.

New 3.5 Flash model is clearly quite optimized for speed. People on this site just look at benchmarks, and underestimate how vital speed is to user experience. No other free model comes close in terms of speed and intelligence, so it seems like the correct strategy to slowly win the mass market, as almost everyone will form their experience of AI through its free version.