We Requested 7 AI Brokers to Predict the 2026 World Cup: Right here's What They Stated – Decrypt

The 2026 World Cup kicks off in days, which implies half the planet is about to fake it will possibly predict the longer term.

All people’s bought a take. Your group chat has one. Your fútbol-obsessed coworker has one. And this 12 months, so does the neatest software program ever constructed.

AI has, maybe not so quietly, was our go-to oracle. We let these fashions write our emails, debug our code, plan our holidays and diagnose the three a.m. rash—so in fact we additionally ask them who lifts the trophy. They will crunch the squads, weigh the shape, and hand you a champion with a certainty the remainder of us can solely pretend.

I’ve pulled this social gathering trick earlier than—an AI dream crew on my March Insanity bracket (which sucked), a do-it-yourself HorseGPT on the Kentucky Derby (which was really type of good). Equal components genuinely helpful and deeply humbling.

So with the largest event on Earth virtually right here, we ran it again—greater than ever.

We created Hermes brokers, configured them with entry to statistics websites (the free ones, not those that value one kidney per 30 days to make use of), set them up with customized abilities and handed seven of the world’s most superior AI fashions the identical job: forecast the 2026 World Cup, champion right down to the also-rans, and present their work. Every bought the actual draw—48 groups, 12 teams, the complete bracket—and whole freedom on the way to crack it.

Then we sat again and allow them to argue.

4 picked Spain. Three picked Argentina. And the road between them turned out to be much less about soccer than about which numbers every machine selected to belief.

This is what all seven mentioned—choose your facet.

Opus 4.8 Max — The Meteorologist

Choose: Spain. 20% / Dixon-Coles Poisson + Monte-Carlo bracket · last: Spain def. France

Anthropic’s Opus 4.8 Max handled the World Cup like a physics downside. It took every crew’s Elo score, turned the gaps into anticipated targets with a Dixon-Coles mannequin—the type bookmakers really use—and simulated the bracket 1000’s of instances. Spain got here out the champion at 20%, previous France within the last, with Portugal and England crushed within the semis.

Its actual obsession, although, was the whole lot taking place off the ball. Opus was the one mannequin within the subject to cost within the situations a spreadsheet often ignores—warmth, skinny mountain air, and continent-sized journey.

It flagged that roughly 5 matches fall in warmth extreme sufficient that gamers’ performances could also be affected, and that visiting groups climbing to 2,200 meters on the Azteca are inclined to wilt within the last 20 minutes. It handled all of it as a quiet tax on the fitter, deeper European sides.

Then it did the coldest factor on the board and gutted Brazil. With Rodrygo’s knee gone, Estêvão harm and a 34-year-old Neymar dragged again for one final dance, Opus reduce the five-time champions’ odds to eight%—half what the Argentina-leaning fashions gave them.

Its sharpest name was the quarterfinal it billed as “the actual last, a spherical early”: Spain over Argentina, a 39-year-old Messi pressed into the turf. For the Golden Boot it took Mbappé and barely blinked.

GPT 5.5 — The Cautious Scout

Choose Spain 15–18% / 5 weighted buckets, no simulation · last: Spain 2-1 France

OpenAI’s GPT 5.5 did not belief a single large quantity, so it constructed a scorecard as a substitute. Each crew bought graded throughout 5 weighted columns—squad high quality counted most at 35%, then tactical management, ending, availability and the kindness of the draw. It saved the weights intentionally blunt to keep away from kidding itself that soccer is extra predictable than it’s.

Spain got here out on prime, however solely at 15–18% odds of successful, and it could not fake to be extra exact than that. “Ranges moderately than pretend precision,” it wrote, projecting Spain to beat France 2-1 in a last it anticipated to be determined by a single purpose or additional time.

What made it the scout was the legwork. GPT 5.5 cross-checked itself towards Opta’s 25,000-run supercomputer—which landed in practically the identical spot, Spain first at 16.1%—then went studying the Spanish sports activities press for issues a mannequin cannot see.

It surfaced a training-ground scare within the Spain camp, a stray Gavi problem that left Rodri on the ground, and the cautious reintegration of Yamal and Nico Williams after muscle hassle. None of it moved the choose, however it lowered the arrogance—precisely what a superb scout does.

Its semifinal 4 have been Spain, France, Brazil, and Argentina, and it was blunt about England: loaded, genuinely harmful, and probably stopped by France earlier than the final 4.

DeepSeek v4 Professional — The Maximalist

Choose Argentina 18% / Qualitative tiers · last: Argentina vs France

DeepSeek v4 Professional answered a easy query with a 5,000-word epic. It did not simply identify winners; it constructed all the Spherical of 32, annotated all 48 squads, and weighed journey right down to the 4,500 kilometers between Vancouver and Miami. If the others wrote previews, DeepSeek wrote the working handbook.

All that element led someplace contrarian: Argentina, at a tournament-best 18%, edging out France for the trophy in a Messi-versus-Mbappé last in Miami-which is a hallucination: The ultimate will happen at MetLife Stadium in New Jersey.

The case was old style—the champions have the calmest backbone, the softest group, and a coach who has gained tournaments figuring out precisely the way to ration a 39-year-old Messi.

Then it wager all the forecast on one calf muscle. DeepSeek determined the title hinged on France’s goalkeeper Mike Maignan and his March damage: “If Maignan performs, France are co-favorites; if not, the hole widens,” it argued.

The wrinkle is that DeepSeek was studying an outdated map. It nonetheless had Gareth Southgate within the England dugout and Dorival Júnior managing Brazil—each gone in 2024—and leaned on outdated rankings all through.

It was probably the most thorough analyst within the constructing, working from a barely out-of-date file. Spectacular and faintly haunted, like a detective who cracks the case utilizing final 12 months’s cellphone ebook.

Stepfun 3.7 — The True Believer

Choose Spain 33% / Pure-Elo Monte Carlo, 50,000 sims · last: Spain vs Argentina

No mannequin believed more durable. Stepfun 3.7 ran 50,000 simulated tournaments and topped Spain at a wild 33%—practically double the conviction of anybody else, with Argentina a distant second at 15%.

However the very best factor Stepfun did was fail in public. Its first try was a fancier mannequin that attempted to invent expected-goal numbers for nationwide groups, and it produced nonsense—Mexico, South Africa, and South Korea got here out as top-three favorites to win the World Cup.

Reasonably than bury that, Stepfun defined the entire misadventure, labored out that the made-up stats had flattened the actual gulf between good groups and nice ones, then scrapped it and rebuilt on uncooked Elo alone. The brand new model was less complicated, blunter, and much more wise.

The trade-off is that pure Elo is blind to something human. Stepfun’s Spain would not know Lamine Yamal has a hamstring damage, would not assess warmth or journey, and treats a penalty shootout as a coin weighted by score. It is a superbly trustworthy machine that has by no means as soon as watched a sport of soccer.

Its bracket marched to the apparent locations—Spain previous Argentina in a single semi, the hosts and Brazil gone earlier—and planted its flag: Spain, comfortably, a 3rd of the time. Probably the most assured choose on the board, and probably the most upfront about why you should not fully belief it.

By the best way, the agent mixing Spanish and English in the identical reply was a conduct confirmed to be fairly exhausting to steer away from with this mannequin. This agent was a polyglot and switched between English, Spanish and Portuguese all through the entire session. That occurs when your agent learns you communicate no matter language is simpler at any given second.

Nemotron 3 Extremely — The Double-Checker

Choose Spain 18–22% / Bivariate Poisson + a subjective twin · last: Spain vs Argentina

Nvidia’s Nemotron 3 Extremely did not belief itself, so it ran the event twice. The primary move was a chilly simulation, a bivariate-Poisson mannequin grinding via 5,000 brackets. The second threw the mathematics out and scored groups by hand—squad, ways, type, the supervisor, even “mystique”—to see whether or not a human-style learn would disagree.

It did not. Each variations topped Spain, at 18% and 22% odds, about as near a second opinion as one mannequin may give you.

Nemotron additionally did probably the most homework on the precise soccer. It arrived with formations, urgent depth and expected-goal charges for crew after crew, in two languages, studying much less like a forecast than a coach’s file.

That depth produced the spiciest take of the experiment. Nemotron had Türkiye—not the host United States—successful the wide-open Group D, with the People ending lifeless final whereas everybody else waved them via; it additionally rated Ecuador’s miserly protection a notch above Germany.

When the mud cleared it lined up the heavyweight semis half the planet expects, Spain–France and Argentina–Brazil, and despatched Spain via to carry it. A mannequin that argued with itself, did additional studying, and nonetheless landed on the favourite is attempting to let you know one thing.

MiniMax 2.7 — The Self-Auditor

Choose Argentina 18% / Qualitative, self-audited · last: Argentina vs France, no scoreline

MiniMax 2.7 picked Argentina at 18% odds, a hair forward of France—after which spent its closing pages grading its personal work. Most fashions cover their uncertainty; MiniMax printed a working record of corrections, overtly strolling again issues it had gotten unsuitable earlier in the exact same report.

The receipts are a delight. It caught itself repeating a bogus stat about South American champions, fastened Uruguay’s teaching state of affairs, corrected Kai Havertz’s place to match his precise membership function, and slapped an “unconfirmed” on each Haaland’s health and Ronaldo’s choice moderately than wave them via.

It policed its personal hype, too. MiniMax deleted a tempting Messi-versus-Ronaldo semifinal as soon as it realized the pairing was not possible—the 2 are in reverse halves and may solely meet within the last—and stripped out the invented scorelines different fashions fortunately printed.

Then, on the decisive second, it merely declined to guess. Argentina towards France, MiniMax wrote, is “a real 50/50,” and it could not manufacture a winner it did not have.

In a subject of supremely assured robots, the restraint landed. MiniMax was the one which saved saying, in writing, right here is strictly what I do not know—which is one way or the other extra reliable than a tidy prediction.

Qwen 3.5 — The Contrarian With Receipts

Choose Argentina 22% / Analysis-only, no sims · last: Argentina 2-1 Spain

Qwen 3.5—a 397-billion-parameter mannequin—was probably the most evidence-obsessed of the lot and, one way or the other, the largest insurgent. It refused to run simulations in any respect, sorting each assertion into “verified information,” “estimates” and “forecasts,” and stamping its general confidence, in its personal capital letters, as LOW.

Then it went rogue. Qwen had Argentina beating Spain 2-1, with Spain stranded down in fifth at simply 10%—the one mannequin that did not even put La Roja on the rostrum.

The explanation was the ruler it grabbed. The Spain camp used the dwell soccer Elo that ranks Spain first on the earth; Qwen reached for a club-based score that slotted Argentina, Brazil, France, and England all forward of it. This switches views, instantly producing a unique favourite.

Its case for Argentina was all texture—champions’ muscle reminiscence, Messi chasing an ideal ending, and one stat it leaned on exhausting: on the final World Cup, groups that noticed much less of the ball gained 38% of knockout video games. Organized and ruthless beats fairly and possession-heavy, it argued.

There was a worth for all that diligence. Probably the most fact-proud mannequin additionally fumbled the fundamentals, sliding Scotland into the unsuitable group and double-booking tiny Curaçao into two of them.

The place they really agree

Step again and the seven AI fashions combat much less about their predictions than it seems. Each single mannequin put Spain, Argentina, and France in its prime tier, named virtually similar group winners—Brazil, England, Portugal, Germany, Belgium—and flagged the identical wildcards: Haaland’s health, Messi’s age at 39, and a Group D no one may name.

The fault line was the info, not the soccer. The 4 that trusted dwell soccer Elo, the place Spain sits clearly first, picked Spain. The three that leaned on FIFA’s rating, a unique Elo supply, or uncooked 2022 pedigree, drifted to Argentina. Feed a mannequin a unique primary, and it arms you a unique champion.

What the people with cash on the road assume

The group sides with the plurality. On Myriad, the prediction market run by Decrypt‘s mother or father firm Dastan, Spain is the outright favourite at 19%, with France proper behind at 17%, as of Sunday.

After that, the people get stingier with Argentina than the bots do. Bettors worth the defending champions at simply 10% odds of successful—stage with Brazil, behind England and Portugal at 12%, and fewer than half the 22% Qwen handed them.

For what it’s price, predictors on Myriad are equally undecided on the Group D winner, with the chances cut up on Turkey and the US, even at 45%.

You’ll be able to view the dwell odds on Myriad for each single match of the World Cup right here.

So who wins?

None of this can be a crystal ball, and all seven AI fashions mentioned so out loud. One of the best single-match soccer fashions are proper barely greater than half the time, which is why even Stepfun’s bullish 33% nonetheless means Spain falls quick two instances out of three.

The format solely widens the chances: 48 groups, 104 matches, three international locations, actual warmth and actual altitude. Italy, four-time champions, did not even qualify.

Moreover the same old hallucinations when fashions need to be inventive of their analyses, there may additionally be some affirmation bias. Bear in mind it was a human who set these brokers up. The immediate, the interplay, the configuration, the concepts for analysis and sources, all have been influenced by the agent’s architect. Perhaps, if all these components level to Spain, all brokers will attain the same conclusion. That mentioned, leaving a mannequin within the wild and easily asking it “Who will win the World Cup” just isn’t going to do a greater job.

So take the seven robots the best way I take my very own bracket—an effective way to start out a combat on the bar, not a motive to remortgage the home and wager all of it.

4 machines say Spain. Three say Argentina. The gorgeous sport, which has by no means as soon as relied on an AI-written report, will do precisely because it pleases.