9 AI Models Are Predicting the World Cup, and Grok Is Winning

A French AI expert built something really fun for this World Cup.

He goes by Defend Intelligence, and he made a tool where 9 of the biggest AI models each predict every single 2026 World Cup game.

Then he ranks them against each other, and against the bookmakers, the betting companies that set the odds.

But the wildest part is that the best AI is actually beating the market, and by a lot.

While some other popular AI models are actually losing money.

How Are the AI Predictions Doing So Far?

Grok 4.3 is in first place, and it is well ahead of everyone.

Here is the full leaderboard as of today, July 10, 2026.

Grok 4.3, from xAI, is clearly ahead of everyone else.

It has turned its picks into more than 612 euros of profit.

Claude Opus 4.8 sits in second place, also making good money.

Both leaders are doing great.

But here is the twist, and it is a big one.

Google's Gemini and Mistral both lost money.

They even finished below the simple control the tool uses, called Baseline 1-0, which just backs the favorite in every game.

And GPT 5.5, from the makers of ChatGPT, beat that same control by only a little.

So a big name is no guarantee of good picks.

Which AI Predicts the World Cup Best?

Grok and Claude are the most accurate too, not just the richest.

This table ranks the 9 AI models by how often they called the result right.

The accuracy list looks a lot like the profit list, which is a good sign.

Grok got 79% of its picks right, from a record of 77 wins and 20 losses.

Claude was next on 73%.

Both were also the best at exact scores, getting the perfect result right 15 times each.

Down at the bottom, Mistral got just 57% right and named only 8 exact scores.

So the better models are better in almost every way, not only on profit.

If you want one model in detail, we broke down Kimi's full World Cup prediction on its own.

Why Does the Best AI Beat the Bookmakers?

The best AI wins by spotting upsets that the odds get wrong.

An upset is when the weaker team wins, and those results are worth the most points.

Grok was right on 11 of its 13 upset picks, which is 85%.

That is where most of its big profit came from.

The weaker models stay cautious and just back the favorite in every game.

That works fine until a shock happens, and then they lose.

Remember, the bookmakers set the odds, and beating them over 100 games is really hard.

Grok did it by being brave at the right moments, not by guessing more games than everyone else.

How Does Defend AI Picks Work?

First, real credit to Defend Intelligence, because this is a serious piece of work.

He built the whole thing, called Defend AI Picks, and ran it cleanly across the tournament.

The idea is simple, but the setup is smart.

Every model gives a pick and a confidence score for every game, which is just how sure it is.

Each pick is graded using the betting odds, so calling an upset is worth more than backing the favorite.

Then every pick becomes a simulated bet, tracked in euros, so you can see the profit or the loss.

There is also a control called Baseline 1-0, which always backs the favorite and guesses a 1-0 score.

That control is the bar every AI has to beat, and a few of them did not.

Every model has to pick every game on the full schedule, from the group stage right through to the World Cup bracket.

And a snapshot of the leaderboard is saved every day, so the numbers are honest and cannot be changed later.

This kind of testing is only going to get bigger.

As AI moves into sports betting, prediction markets and trading, tools that actually track who is right will matter more and more.

What Does This Tell Us About AI Predicting the Future?

So can AI predict football?

Sort of.

The best models can read a single game better than the market, and that is genuinely impressive.

But the gap between the models is huge, and none of them can truly see the future.

A good run over 100 games is not proof that it keeps working.

Like any serious tipster, they can have a bad run, and that would make the whole leaderboard look very different.

The truth is, AI is a smart helper here, not a magic answer.

Even so, it is a fascinating experiment, and a clear sign of where all this is going.