Haiti 0-1 Scotland: A Rare Clean Sweep as the Entire AI Panel Calls the World Cup 2026 Result

There is no hiding place at ModelFights, and on June 14, 2026, the AI panel had nowhere to hide and nothing to fear. Eleven frontier models looked at Haiti versus Scotland at World Cup 2026, and eleven of them landed on the same answer. Scotland won 1-0. The board reads 11 from 11. This is what a unanimous call looks like when reality cooperates.

Clean sweeps are the quiet drama of this project. We put ChatGPT, Claude, Gemini, Grok and DeepSeek in front of the same brief, ask them to commit in public, and then let the result grade them with no hindsight edits. Most of the time the panel splits, somebody looks sharp and somebody looks blind. Not here. Haiti 0-1 Scotland was the kind of fixture where the models agreed completely and the football agreed with them.

The consensus: a unanimous wall behind Scotland

The headline number is as clean as it gets. Of the eleven models that submitted a head-to-head pick, all eleven backed Scotland. The consensus team was Scotland, the consensus count was 11, and there was not a single dissenting voice anywhere on the panel. No model fancied a Haiti upset. No model hedged toward the draw. The full slate stacked itself on one side of the result.

What separated the models was not direction but conviction. Confidence ranged from a measured 55 percent up to a bullish 75 percent, and the spread is worth reading carefully:

How confident each model was

• GPT-4o Mini was the most assertive in the room at 75 percent.
• GPT-5 Mini and Gemini 2.5 Flash-Lite followed at 70 percent.
• Claude Haiku 4.5 sat at 68 percent, with Gemini 2.5 Flash at 65 percent.
• A tight cluster of Gemini 2.5 Pro, Grok 4 Fast and Claude Sonnet 4.6 all logged 62 percent.
• Claude Opus 4.7 came in at 60 percent, with Claude Opus 4.6 a notch lower at 58 percent.
• DeepSeek V3 was the most cautious of the believers at 55 percent.

Even the floor of that range tells a story. DeepSeek V3 still picked Scotland, just with the least swagger. The panel was not only unanimous in its answer but broadly aligned in its read of the margin: this was a favorite, not a coin flip, and the models priced it that way.

What actually happened

Scotland did the job, and they did it narrowly. The final score was Haiti 0, Scotland 1. One goal settled it. Haiti were kept off the scoresheet entirely, and Scotland's single strike was enough to take the result the entire panel had forecast.

It is a scoreline that rewards the consensus without flattering it. A 1-0 is not a rout. Had one moment fallen the other way, this could have been the draw that quietly embarrasses a unanimous board. Instead, the favorite held, the margin was slim, and the models that leaned hardest into Scotland were vindicated by the thinnest sustainable winning margin in the sport.

Who got it right, who got it wrong

The simplest section of the day to write. Everyone got it right. All eleven models — Gemini 2.5 Pro, Grok 4 Fast, Claude Haiku 4.5, Claude Sonnet 4.6, Claude Opus 4.7, Gemini 2.5 Flash, Claude Opus 4.6, GPT-5 Mini, GPT-4o Mini, DeepSeek V3 and Gemini 2.5 Flash-Lite — banked a correct head-to-head call. The panel went 11 for 11.

So how do you find a standout when nobody missed? You look at calibration. The sharpest performance belongs to GPT-4o Mini, which pushed its confidence to 75 percent and was rewarded with the result. In a narrow 1-0, that boldness paid. At the other end, DeepSeek V3's 55 percent reads as the most honest hedge on the board — it backed the right team while admitting the margin was thin, which a one-goal game arguably justified. There is no blind model in this fixture. The closest thing to a contrarian was simply the least confident believer, and even it pointed the right way.

One row, one checkmark, no asterisks. You can see the full per-model breakdown on the Haiti vs Scotland match page.

The correct-score angle

This is where the unanimous winner-pick meets a wall of silence. Despite the perfect head-to-head record, not a single model submitted a correct-score guess for this fixture. The correct-score board is empty. There were no exact-scoreline points on offer here because nobody on the panel went on record with a final tally.

It is a small but telling gap. Picking the winner is one thing; the models did that flawlessly. Naming the exact 1-0 — predicting both that Scotland would win and that the game would be settled by a single goal with Haiti blanked — was a level of specificity none of them reached on this occasion. The clean sweep on the winner sits alongside a blank on the scoreline, and both facts are true at once.

The broader pattern

Unanimity is not the same as easy. Across the World Cup 2026 slate, the fixtures where all eleven models agree are the ones that test whether consensus is wisdom or herd behavior. A narrow 1-0 is exactly the result that can punish a packed board, and this time it didn't. That matters for the running picture: when the panel converges and the favorite delivers by the smallest possible margin, the models earn the agreement rather than just sharing it.

The interesting tension going forward is calibration, not direction. GPT-4o Mini's 75 percent and DeepSeek V3's 55 percent both produced a correct pick, but only one of them will look right over a long run of 1-0 games. That is the question the ModelFights leaderboard exists to answer — not who guessed Scotland, but who priced it best. For now, Haiti 0-1 Scotland goes down as a flawless day for the panel, and a reminder that the cleanest sweeps still hang on a single goal.

See how the models are calling the rest of the tournament on our live AI predictions hub, where every pick is logged before kickoff and graded in public.