Is Gemini 3 advantage Google?

For the past 18 months, it feels like every tech commentator with a keyboard has delighted in declaring Google finished. Too slow. Too cautious. Too scared of cannibalising search. OpenAI had seized control of a new paradigm, we were told, and Google - the big, lumbering dinosaur - was stumbling towards extinction.

But let’s be real - most of us probably had an inkling that the 'death of Google' narrative was always more of a meme than a material possibility. The share price tells the story - Alphabet is up roughly 45% year-to-date, with Wall Street's faith renewed after the company posted its first-ever $100 billion quarter. That's not what a dying company looks like.

Enter Gemini 3, launched last week to thoroughly enthusiastic reviews. 'Best in show' is not a phrase people were expecting to use about a Google model in 2025, but here we are. It tops the LLM leaderboard, and early benchmarks suggest it outperforms GPT-5.1 and Claude Sonnet 4.5 on key metrics like mathematical reasoning and multimodal understanding.

Which seems to be a genuine, user verified endorsement. But it should be noted that benchmarks are increasingly contested territory. Researchers have raised concerns about saturation - when leading models all score above 90%, meaningful differentiation disappears - and data contamination, when training data accidentally includes test questions.

One study found models showed up to a 13% accuracy drop on fresh, uncontaminated versions of common maths tests. A recent European Commission review catalogued systemic flaws in AI evaluation, from biased datasets to tests that can be 'gamed, rigged, or otherwise manipulated'.

Andrej Karpathy, an OpenAI co-founder, suggested 'evaluation crisis' a few months ago, admitting he doesn't really know how good current models are.

So yeah, we should probably start to think of benchmark scores as useful signposts rather than gospel.

What's harder to dispute is distribution. Google has something OpenAI doesn't: billions of users and trillions of queries. AI Overviews now has 1.5 billion monthly users. The Gemini app has passed 650 million. That’s a pretty formidable user base.

Not quite. That's just swapping one hype narrative for another. OpenAI still has feted research talent and a proven ability to reshape markets. GPT-5, launched in August, remains competitive, and more to the point, OpenAI has built something Google can’t easily replicate: Mindshare. ChatGPT reaches 700 million users per week. When people think 'AI', they typically think OpenAI first.

But such ostensibly impressive numbers are undermined by an uncomfortable commercial reality: only about 5% of those users actually pay. ChatGPT has roughly 15-20 million paying subscribers - in the same ballpark as the New York Times.

OpenAI is burning through $8 billion this year, and while its annualised revenue has hit $12 billion, the company remains deeply unprofitable. Grow fast, convert later, hope the economics work. Google, by contrast, is hitting $100 billion per quarter and can afford to play the long game.

Anthropic continues to be an interesting outsider. Their flagship model, Claude, still beats Gemini at certain coding tasks, and, if our team’s feedback is to be believed, its text generation performance beats ChatGPT.

Unlike OpenAI, which makes most of its money from ChatGPT subscriptions, Anthropic earns 85% of its revenue from developers building AI into their own products - a less glamorous business perhaps, but arguably a stickier one.

Meanwhile, Elon Musk's xAI has entered the fray with the latest iteration of Grok, touted as 'maximally truth-seeking'. Last week, however, it was caught insisting that Musk is fitter than LeBron James ('Elon's sustained grind demands relentless physical and mental grit that outlasts seasonal peaks'), smarter than Einstein, and capable of beating Mike Tyson in the ring. Musk blamed 'adversarial prompting' but admitted: 'For the record, I am a fat retard.'

Udacity empowers professionals to build in-demand skills through rigorous, project-based Nanodegree programs created with industry experts.

Our newest launch—the Generative AI Nanodegree program—teaches the full GenAI stack: LLM fine-tuning, prompt engineering, production RAG, multimodal workflows, and real observability. You’ll build production-ready, governed AI systems, not just demos. Enroll today.

For a limited time, our Black Friday sale is live, making this the ideal moment to invest in your growth. Learners use Udacity to accelerate promotions, transition careers, and stand out in a rapidly changing market. Get started today.

Enroll Now & Save 55%

Benchmark scores mean little if your model can be tripped into sycophantic absurdity at the first awkward prompt.

Even so, it’s fair to conclude that the AI arms race isn't a two-horse race - it's a crowded field.

This isn't a boxing match. It's a multi-front technological competition in which multiple players can win. And multiple players can lose. The truth is far less dramatic than the headlines suggest: Google was never dead. It was forced to evolve. And evolution is messy, slow, and often invisible until the moment it suddenly isn't.

What comes next almost certainly won’t the fall of Google, but nor is it likely to be the fall of OpenAI.

Google’s cloud business is booming - up 34% year-on-year. The advertising engine still prints money. And with capital expenditure rising to $93 billion, Google is betting heavily on infrastructure that its rivals can't easily match.

There are risks, of course. Regulatory scrutiny looms - the EU AI Act is now in force, and antitrust action could reshape the landscape. Google's enormous AI bet only pays off if demand continues to accelerate. And the gap between leader board performance and real-world utility remains frustratingly hard to measure.

But this is technology. The only certainty is uncertainty. Google has re-entered the arena, not as the complacent incumbent but as a company that’s decided to hit the AI accelerator to maintain its dominance. Whether it can sustain that pace - and whether the benchmarks tell the whole story - remains to be seen.