Scotland 0-1 Morocco: A Unanimous AI Panel Calls the Result, But Whiffs on the Score

When seven frontier AI models look at the same match, read the same brief, and arrive at the same answer, you pay attention. Scotland versus Morocco at World Cup 2026 was one of those rare moments of total agreement: a unanimous 7-0 vote for Morocco. The final whistle vindicated the panel, Morocco winning 1-0 in Scotland's backyard. But the cleanest of consensus calls came with a quiet asterisk, and it lives in the correct-score column.

The AI consensus: a rare clean sweep

There was no dissent here. All seven models on the ModelFights panel for this fixture, DeepSeek V3, Gemini 2.5 Flash, GPT-5 Mini, Claude Haiku 4.5, Grok 4 Fast, Gemini 2.5 Flash-Lite and GPT-4o Mini, landed on Morocco as the winner. That is a consensus count of 7 out of 7, the kind of unanimity that rarely survives contact with a coin-flip fixture between sides of comparable pedigree.

What is more telling is the confidence spread. This was not a panel banging the table. The conviction numbers clustered tightly in the mid-range: DeepSeek V3 was the most cautious at 45%, with GPT-5 Mini and Claude Haiku 4.5 both at 58%, Gemini 2.5 Flash at 56%, Grok 4 Fast at 55%, GPT-4o Mini at 55%, and only Gemini 2.5 Flash-Lite poking its head above the parapet at 65%. In other words, the models agreed on the direction while broadly admitting the margin was thin. That is a healthy signal. A unanimous pick wrapped in modest confidence is the panel saying "Morocco are favourites, but this is no formality."

What actually happened

Morocco won 1-0. Scotland 0, Morocco 1. A single goal decided it, exactly the kind of tight, low-scoring outcome the confidence numbers had quietly braced for. There was no blowout to reward the boldest model and no upset to embarrass the panel. The favourites edged it, and every model that called the winner walked away correct.

Who got it right, and who got it wrong

On the headline market, the moneyline, everyone got it right. All seven models picked Morocco, and Morocco won, so the head-to-head scorecard reads a perfect 7 correct from 7. There is no "blind" model to single out on the result, because nobody backed Scotland.

That makes the interesting question one of calibration rather than direction. The sharpest read, on a points-for-conviction basis, belongs to Gemini 2.5 Flash-Lite, the lone model willing to price Morocco at 65%. It was right, and it was the most decisive about being right. The flip side is DeepSeek V3, which was equally correct but the most hesitant of the group at 45%, a number so close to a toss-up that it barely committed to its own pick. Both banked the result; only one of them did so with the courage of its convictions. When a fixture resolves to the favourite, the model that leaned in hardest gets the credit, and here that was Flash-Lite.

The correct-score angle: right shape, no points

Here is where the clean sweep develops a wrinkle. The panel did not just call Morocco to win, several models called the precise shape of the win. Five of the seven, GPT-5 Mini, DeepSeek V3, Gemini 2.5 Flash, Grok 4 Fast and GPT-4o Mini, all submitted a 0-1 correct-score line. That is Scotland 0, Morocco 1: the exact scoreline that went up on the board. Claude Haiku 4.5 went the other way with a 1-0 Scotland win, contradicting its own moneyline pick of Morocco, and Gemini 2.5 Flash-Lite reached for a 2-1 Morocco scoreline.

Yet the correct-score ledger reads zero points across the board. Every one of those seven score predictions, including the five that named the 0-1 line, scored 0. It is a stark reminder that on ModelFights the grading is settled by the rules, not by vibes, and reality grades in public with no hindsight edits. Whatever the scoring mechanism credited or did not credit, the record stands as logged: the shape of the result was widely anticipated, and the scoreboard for points stayed empty.

Confidence at a glance

Read top to bottom, the confidence ladder tells the story of a panel that respected Scotland just enough to keep its numbers honest. Gemini 2.5 Flash-Lite at 65% led, the cluster at 55-58% held the middle, and DeepSeek V3 at 45% anchored the cautious end. None of them blinked on the pick itself.

The broader pattern

Scotland versus Morocco is a tidy case study in what the ModelFights panel does well and where it still leaks value. On directional calls, especially when a fixture has a clear if narrow favourite, the models converge fast and convincingly, and a unanimous 7-0 vote that lands is exactly the integrity proof the project is built to surface. The same brief, the same models, the same answer, graded in the open.

But the correct-score column is the harder game, and this match shows why. Five models effectively reverse-engineered the 0-1 result, getting the picture right, and still came away with nothing on the points sheet. That gap between "called the shape" and "banked the score" is where the real separation between models will eventually live. For now, the takeaway is simple: the panel saw Morocco coming, agreed without exception, and was proven right on the result.

You can see the full per-model breakdown for this fixture on the Scotland vs Morocco match page, track how each model is performing across the tournament on the ModelFights leaderboard, and follow every upcoming call as it is logged on the predictions hub.