Why does ScoreGPT use five AI models?

GPT-6.1 Sol, Claude Opus 5.5, Grok 4.7, DeepSeek-V4-Pro and Kimi K3.1 Pro. Each one comes from a different lab, and each one researches the match with live web search — team news, injuries, form — before it makes its call.

Model	Lab
GPT-6.1 Sol	OpenAI
Claude Opus 5.5	Anthropic
Grok 4.7	xAI
DeepSeek-V4-Pro	DeepSeek
Kimi K3.1 Pro	Moonshot AI

None of them sees another model's answer before it makes its own. Each has its own page with its record: the five models.

• Each model's own call, side by side — on the match pages here and in the app. The app adds each model's written analysis.
• Where they agree and where they split — five out of five and three against two never look alike.
• One combined call — the five models' chances averaged into one call. It is the five combined, not a sixth model.
• Every result afterwards — each call is checked against the final score and kept, wins and misses. See the results.

Every call is a full research run. A frontier model searches the latest news on the match, reads it and reasons through it before it answers, and that costs real money for every model on every match. So we chose depth over head-count: five models that each do the whole job, with search, on every match we cover. More models would mean thinner research per model, or fewer matches.

Every model's record, and the combined call's, is public match by match — see the results.

Football stays uncertain. A red card, a deflection or a goalkeeper in the form of his life can turn any match. No model gets every match right, the combined call included, and ScoreGPT makes no guarantees: a prediction is an opinion about an uncertain event, never a guarantee of the result. That is why every miss stays on the record, next to every hit.