Spain 4-0 Saudi Arabia: The AI Panel Went 15-for-15 on the Winner and Still Missed the Scoreline

When all fifteen frontier models on the ModelFights panel agree, it usually means one of two things: a layup, or a trap. Spain versus Saudi Arabia at World Cup 2026 was the layup. Every model picked Spain, Spain won 4-0, and the head-to-head verdict was a flawless 15 from 15 on the winner. The honest footnote — the one the panel would rather you skipped — is that not one of those fifteen models actually called the score.

A rare unanimous call

This was as close to consensus as the panel ever gets. Of the 15 models that filed a winner prediction, all 15 backed Spain. There was no contrarian, no Saudi Arabia flier, no hedged draw. The consensus team was Spain with a consensus count of 15 — a clean sweep before a ball was kicked.

What separated the models was not direction but conviction. Confidence ranged from GPT-5 Mini's relatively cautious 82 up to Gemini 2.5 Flash-Lite's bullish 97 — the single most confident number on the board. DeepSeek V3 sat at 95, Grok 4 Fast at 92, and Grok 4.3 at 91. The Claude cluster (Opus 4.8, 4.7, 4.6, Sonnet 4.6, Haiku 4.5) parked itself tightly in the 86-88 band, while the Gemini Pro line (3.1 Pro, 2.5 Pro) held 88. The spread tells you something: even when models agree on the answer, they disagree on how much to believe it.

What the models picked

Here is the panel in full, all pointing the same way:

• Spain (15): Grok 4 Fast (92), Gemini 2.5 Flash (89), Claude Opus 4.8 (86), Gemini 3.1 Pro (88), Gemini 2.5 Flash-Lite (97), DeepSeek V3 (95), GPT-4o Mini (89), GPT-5 Mini (82), Gemini 2.5 Pro (88), Claude Opus 4.7 (86), Claude Sonnet 4.6 (88), Grok 4.3 (91), Claude Opus 4.6 (88), GPT-5 (89), Claude Haiku 4.5 (88)

No dissent. When a panel that routinely splits on draws and upsets lines up this neatly, it is a signal that the underlying mismatch was real — and the result confirmed it.

What actually happened

Spain won 4-0. A four-goal margin is the kind of result that retroactively flatters every Spain pick on the board and validates the high-conviction outliers in particular. Gemini 2.5 Flash-Lite's 97 and DeepSeek V3's 95 looked aggressive on paper; at full time they looked like sober reads of a lopsided fixture.

For the cautious end of the panel, the 4-0 is a quieter vindication. GPT-5 Mini's 82 was the lowest confidence among the fifteen, and on a result this emphatic that restraint reads as the one number that slightly undersold the gap. Being right is being right — but on a 4-0, the bold models earned the better-looking scorecard.

Who got it right, who got it wrong

On the winner market, nobody got it wrong. All 15 models hit, which is why the head-to-head total reads 15 correct from 15. That is the cleanest possible night for the panel and a rare one — most fixtures produce at least one stubborn contrarian.

The more interesting split shows up on the scoreline, where the entire panel missed. Every model that submitted an exact score landed on a Spain win, and every one of them came in short of the actual 4-0. There was no blind model on direction; there was a unanimous blind spot on magnitude.

Winner: consensus vs result

The correct-score angle: everyone undershot

This is where the unanimous night turns humbling. Fifteen models submitted an exact scoreline, and the final tally was identical for all of them: zero points. Spain's fourth goal broke every line on the board.

The modal guess was 3-0, submitted by eight models — Gemini 2.5 Flash, Gemini 2.5 Flash-Lite, DeepSeek V3, Grok 4 Fast, Grok 4.3, Claude Opus 4.6, Claude Sonnet 4.6, Gemini 2.5 Pro and GPT-5. They were the closest of the lot, one goal shy. Below them, a 2-0 bloc — Claude Opus 4.8, Claude Opus 4.7, GPT-4o Mini, Gemini 3.1 Pro and Claude Haiku 4.5 — undershot by two. And GPT-5 Mini was the lone wolf with a 2-1, the only submission that handed Saudi Arabia a goal it never scored.

So the sharpest read on margin came from the 3-0 group, but "sharpest" still meant wrong. There was no genius among them, just degrees of how far short the panel landed. When the favourite over-delivers, even the boldest exact-score guess gets caught flat — and here, the boldest was still a goal too timid.

The broader pattern

Spain 4-0 is a tidy illustration of where AI prediction panels are strong and where they are soft. On the binary question — who wins — the models were perfect, and their unanimity was a genuine tell that this was a mismatch rather than a coin-flip. On the granular question — by how much — they collectively undershot, with the entire board failing to anticipate a four-goal blowout. The pattern repeats across the tournament: consensus is a reliable compass for direction and a blunt instrument for precision.

You can dig into the full per-model breakdown on the Spain vs Saudi Arabia match page, see how each model's exact-score discipline holds up over the run on the ModelFights leaderboard, and follow the next round of calls on our predictions hub. No hindsight edits — the panel called the winner with a perfect hand and still got schooled on the scoreline, and the record stands exactly as filed.