Nobody’s winning the AI race yet

By Ryan Swift

Yesterday, Google finally released its Gemini 4 Argon model for public use. It’s Google’s first flagship release since Gemini 3.1 Pro in February. Whether or not real-world usage lives up to the promising evals, reports of Gemini’s death have been greatly exaggerated. We’re still incredibly early in AI’s development, and while the pace keeps accelerating, there’s plenty of room for comebacks and setbacks in this race.

Over the summer, Google shipped new Flash models at a steady clip, and I found each one was noticeably better than the last. You wouldn’t have known it from the online commentary, which treated Google’s entire model family as unusable. At the AI Engineer World’s Fair conference in July, I wrote that most tasks don’t need a frontier model. All summer, Gemini Flash and Flash-Lite quietly handled the vast majority of my real work. Now that Argon is out, I’d bet many of those same voices will flip-flop over the next few weeks and declare Google destined to win. That’s just as wrong.

It’s still far too early to declare winners and losers of the AI model race. First-mover advantage doesn’t seem to carry the weight it once did. Google was the first major lab in the always-on assistant space (Gemini Spark), but it has since been leapfrogged by Grok Bot, Meta Muse, and OpenAI’s Dots. There’s no clear winner here either. All of these products (and their underlying models) are still in the earliest stages of adoption. Until my mom asks me which assistant to use, they haven’t hit the mainstream.

If anyone’s telling you who’ll win the race, they’re just guessing. Stay flexible. Try the latest releases. Build a small set of your own evals and tasks, because your benchmark matters more than the leaderboard. Avoid locking yourself into any one model. And give Gemini 4 a shot. I bet it’ll surprise you.