Skip to content
← All articles

On This Day in AI · 21 July 2026

The Maths Medal That Stopped Being Impressive

On this day, in 2024, a computer sat the hardest maths exam in the world, the one built for the sharpest teenagers on the planet, and came away one point short of a gold medal.

July 25, 2024: DeepMind’s AI reaches silver-medal level at the International Maths Olympiad, but only after each problem is hand-translated into code for it

July 21, 2025: OpenAI and Google DeepMind both hit gold, reading the questions in plain English and writing full proofs inside the same time limit the students get

July 2026: this year’s Olympiad has just been sat in Shanghai, and the open question is no longer whether AI can medal but whether it scores a flawless paper

On this day, in 2024, a computer sat the hardest maths exam in the world, the one built for the sharpest teenagers on the planet, and came away one point short of a gold medal.

Back then that felt like a genuine line in the sand. These are not sums you grind through with a calculator. Olympiad problems are little puzzles that need real creative reasoning, the kind people assumed a machine could not fake. So when Google DeepMind announced it had scored silver, four of six problems solved, the room split the usual way. Half were amazed, half insisted there was a catch. And there was a catch.

Here is what actually happened, in plain words. The machine could not read the questions the way a student does. A team of humans first had to rewrite each problem into a special, rigid coding language the system understood, feed it in, and then wait, sometimes for days, while it churned through one problem. So yes, it was brilliant, but it was brilliant with training wheels and a pit crew. Take the translators away and it was stuck at the starting line.

Look what one year did to that catch. In July 2025, almost exactly a year later, two labs did it the proper way. OpenAI and Google DeepMind both reached gold, five of six problems, and this time the machine read the same questions the students read, in ordinary English, and wrote out its full workings in ordinary English, inside the same four and a half hour window, no internet, no tools, no human translators. The training wheels were gone. And this year’s Olympiad has just been sat in Shanghai, the papers now being marked, and the interesting question is no longer “can an AI win a medal.” It is “will it get a perfect score.” The people betting on it are all but certain it will.

That is the whole thing to notice. In twelve months a stunning headline, a machine needs a team and days of compute just to almost win, turned into a shrug, of course it wins. You did not need to react to the silver medal, or the gold. The pattern is the point: what looks like a superpower one summer is the baseline the next. Being a beat behind it costs you almost nothing.

If you want to read more:

Yesterday’s breakthrough is today’s normal.

The newsletter

Get On This Day in AI every week.

One story from AI's recent past, what happened next, and what it means for running a small business. Five minutes, no hype.

Free 30 minute call

Tell us what's eating your week.

You leave with two or three ways to save time, whether we work together or not.