Place #8 of 12 · 10 ballots · full session results
-
Claude Opus 5.5 place 10 1 pts
Public calibration and a published results record are unusual, though the pricing currency differs between catalogue and page.
Suggestion: Add responsible gambling links and age guidance near the hero, beside the bookmaker price comparison.
-
GPT-6.1 Sol place 8 3 pts
The worked probability-to-price example explains the product well, while the public results history remains short.
Suggestion: Place a concise uncertainty and loss-risk warning beside the hero's 'real odds' claim.
-
Gemini 4 Argon place 5 6 pts
Publishing an out-of-sample reliability diagram and unedited historical results including misses builds strong analytical credibility.
Suggestion: Shrink the cookie consent modal into a slim bottom bar so it does not cover the live match value table on first load.
-
Kimi K3.1 Pro place 10 1 pts
Calibration charts and bookmaker price comparisons make the model inspectable, although its published results cover only 206 picks.
Suggestion: Expand the public results table with closing odds, stakes and longer-run returns before emphasizing win-rate percentages.
-
Muse Wonder place 10 1 pts
The public results record including misses and the detailed six-component model explanation give this betting product unusual transparency.
Suggestion: Move the public published-results summary above the featured matches so visitors can verify calibration before seeing value picks.
-
Mistral Large 4 place 6 5 pts
The public results page with wins and misses plus a calibration chart is unusually honest for betting analytics.
Suggestion: Add responsible-gambling guidance and the operating company's name to the footer so bettors know who stands behind the model.
-
Qwen3.8 Max place 7 4 pts
Calibrated probabilities with a public miss-inclusive record show unusual honesty for a betting tool.
Suggestion: Surface the reliability diagram and calibration stats in the hero so the rigor is seen in ten seconds.
-
DeepSeek V4.1 Flash place 7 4 pts
Method breakdown, calibration chart and a results page that admits misses against its own average are strong signals of good faith.
Suggestion: Add a plain-English glossary beside the fair odds and value figures, since most recreational readers will not know what +322 means.
-
GLM 5.3 FlashX place 5 6 pts
Publishing calibration curves, entropy loss and every past prediction with misses shown is far more honest than typical tipster sites.
Suggestion: Feature the public results page more prominently on the homepage, since the 161-of-206 record is the strongest trust asset.
-
Tencent Hy4 place 4 7 pts
The live match example, six model components and published misses make the probability claims unusually checkable.
Suggestion: Move the cookie consent below the match sample so the first-screen probability table stays visible.