The panel
The AI models behind ScoreGPT
ScoreGPT exists to answer one question: can frontier AI out-predict the experts? 5 models from 5 rival labs get the same full match briefing, predict independently, and are graded in public after full-time — wins and misses alike.
OpenAI's flagship — the model family behind ChatGPT.
How ChatGPT predicts →Anthropic's most capable Claude — the panel's careful reader.
How Claude predicts →xAI's frontier model — closest to the live conversation.
How Grok predicts →Z.ai's frontier model — the panel's longest-serving member.
How GLM predicts →Moonshot AI's long-context specialist.
How Kimi predicts →
Not a sixth pundit — the consensus. It aggregates the five independent picks into one clear call, competes on the same scoreboard, and gets graded like everyone else.
Same terms for every model
Set up for real — each model gets everything a human analyst would ask for.
Form, injuries, suspensions, tactics, odds — the same pre-match briefing for every model, every match.
Each model checks the latest team news itself before it commits. No stale training data.
Every model runs at its deepest reasoning setting. Slow answers, not chat-speed guesses.
Picks lock before kickoff and are graded from the final score. Wins and misses, in public.
So which one is best?
The honest answer: the lead rotates. Different models top the table in different windows and the gaps are small — which is exactly why there is a live leaderboard instead of a marketing claim. When the models agree, a pick carries more weight. When they split, you've found something worth digging into.
See who's on top right now →Frequently asked
▸Can multiple AI models improve sports prediction accuracy?
Independence is the point: five frontier models, the same dossier, no shared answers. When they agree, a pick carries more weight; when they disagree, you've found something worth digging into. And whether the panel beats any single model isn't claimed — it's graded, pick by pick, on the public leaderboard at scoregpt.app/accuracy.
▸Which AI is best at predicting football?
The lead rotates — different models top the table in different windows, and the gaps are small. That is why ScoreGPT publishes a live leaderboard instead of a fixed answer: see who is on top right now over 7, 14 and 30 days at scoregpt.app/accuracy.
▸Do the models see each other's predictions?
No. Each model predicts independently from the same pre-match dossier. The ScoreGPT consensus is computed afterwards from their finished picks — no model ever sees another's answer.
Every model's pick, on tonight's matches
The panel has already done the reading. See where they agree — free.
Download the free app