What the numbers say

OpenAIs beste KI gets the outcome right in 44% of matches and beats the rule of thumb (36%).

Season 2025/26 · after matchday 34

The AI table

Models, panel and rule of thumb scored by the same system. Tap a row to highlight it everywhere on this page.

Pos.	Model	Picks	Exact	Outcome	Avg points	Index
1	gpt-4o	59	4 · 7%	44%	-0.03	37.4after matchday 34
2	grok-3	57	3 · 5%	44%	-0.26	36.6after matchday 34
3	claude-sonnet-4-5-20250929	32	2 · 6%	44%	-0.14	36after matchday 32
4	gemini-3.1-flash-lite-preview	59	4 · 7%	42%	-0.47	34.3after matchday 34
–	Table favourite wins 1:0	59	4 · 7%	36%	+0.11	–no index

Form = average points per matchday, most recent on the right.strong (+1.5 and up)positivenegativeweak (below −1.5)no pick

The rule of thumb always picks 1:0 for the table favourite. It is the zero line and gets no index, because it states neither confidence nor reasons.

Profiles

Every participant checked against the same questions, so they compare directly.

1.

37Index

stable

after matchday 34

Points form30 · 50 %
Calibration68 · 25 %
Reasoning22 · 25 %

Gets the outcome right in 44% of matches and is overconfident doing so (states 58% confidence on average). Backs the table favourite in 46% of matches. No underdog pick came off.

Outcome right44% · 26/59

Exact scores4 · 7%

Wrong outcomes36%

Stated confidence58%

Bar = actually right (44%), tick = stated confidence (58%).

Reasons confirmed20%

At least one factor could be checked in 71% of predictions.

Most cited reasons

Heimvorteil
cited 28× · 79% confirmed
Keine H2H-Daten
cited 8× · not checkable
Fehlende Formdaten
cited 4× · not checkable

2.

37Index

stable

after matchday 34

Points form29 · 50 %
Calibration73 · 25 %
Reasoning15 · 25 %

Gets the outcome right in 44% of matches and is overconfident doing so (states 61% confidence on average). Backs the table favourite in 51% of matches. No underdog pick came off.

Outcome right44% · 25/57

Exact scores3 · 5%

Wrong outcomes37%

Stated confidence61%

Bar = actually right (44%), tick = stated confidence (61%).

Reasons confirmed14%

At least one factor could be checked in 51% of predictions.

Most cited reasons

Unklare Tabellensituation
cited 7× · not checkable
Neutraler Heimvorteil
cited 6× · 100% confirmed
Fehlende Daten zur aktuellen Form
cited 6× · not checkable

3.

36Index

stable

after matchday 32

Points form30 · 50 %
Calibration66 · 25 %
Reasoning18 · 25 %

Gets the outcome right in 44% of matches and judges its own confidence realistically. Backs the table favourite in 25% of matches. No underdog pick came off.

Outcome right44% · 14/32

Exact scores2 · 6%

Wrong outcomes38%

Stated confidence46%

Bar = actually right (44%), tick = stated confidence (46%).

Reasons confirmed18%

At least one factor could be checked in 78% of predictions.

Most cited reasons

Keine statistischen Daten verfügbar
cited 12× · not checkable
Hohe Prognoseunsicherheit
cited 6× · not checkable
Kein direkter Vergleich vorhanden
cited 5× · not checkable

4.

34Index

stable

after matchday 34

Points form28 · 50 %
Calibration69 · 25 %
Reasoning12 · 25 %

Gets the outcome right in 42% of matches and is overconfident doing so (states 58% confidence on average). Backs the table favourite in 34% of matches. No underdog pick came off.

Outcome right42% · 25/59

Exact scores4 · 7%

Wrong outcomes39%

Stated confidence58%

Bar = actually right (42%), tick = stated confidence (58%).

Reasons confirmed9%

At least one factor could be checked in 46% of predictions.

Most cited reasons

Keine verfügbaren Leistungsdaten
cited 3× · not checkable
Ausgeglichene Ausgangslage
cited 3× · not checkable
Fehlende H2H-Historie
cited 3× · not checkable

Season view

Four angles on the same picks.

Index over time

The index (0–100) weights the last 3–8 matchdays, recent ones count more. Higher is better.

Hit profile

Share of each rung of the scoring ladder across all scored picks of a model.

OpenAI
10%
15%
12%
20%
36%
59
xAI
11%
11%
18%
19%
37%
57
Anthropic
22%
13%
19%
38%
32
Google
20%
19%
39%
59
Rule
12%
10%
20%
41%
59

Exact score+5
Goal-difference hit+4
Partial hit+3
Outcome hit+2
Near miss-1
Miss-2
Bad miss-3

Self-assessment

How often was a model right when it said it was X% sure? Above the line: more cautious than needed. Below: overconfident.

↔ stated confidence · ↕ actually right perfectly calibrated

Easy or hard-earned?

Hit rate when backing the table favourite – and when going against it (underdog or draw).

OpenAIs beste KI

Favourite picked41% · 27 picks

Against the favourite47% · 32 picks

xAI Grok

Favourite picked38% · 29 picks

Against the favourite50% · 28 picks

Anthropics beste KI

Favourite picked50% · 8 picks

Against the favourite42% · 24 picks

Googles beste KI

Favourite picked40% · 20 picks

Against the favourite44% · 39 picks

Rule of thumb

Favourite picked38% · 37 picks

Against the favourite32% · 22 picks

Matchday log

Every match, every pick, every score – ready to check. Tap a pick for the breakdown.

Day ranking
1.xAI GrokØ -1.28 · 3/9
2.Googles beste KIØ -1.35 · 3/9
3.OpenAIs beste KIØ -1.80 · 2/9

Match

OpenAI

xAI

Anthropic

Google

Rize
2:2Fri, 05/15, 05:00 PM
Beşiktaş
no pick
Rize
2:2
Beşiktaş
OpenAI
xAI
Anthropic
no pick
Google
Fatih Karagümrük
2:1Sat, 05/16, 02:00 PM
Alanya
no pick
Fatih Karagümrük
2:1
Alanya
OpenAI
xAI
Anthropic
no pick
Google
Samsun
3:0Sat, 05/16, 05:00 PM
Göztepe
no pick
Samsun
3:0
Göztepe
OpenAI
xAI
Anthropic
no pick
Google
Gaziantep
1:2Sat, 05/16, 05:00 PM
Başakşehir
no pick
Gaziantep
1:2
Başakşehir
OpenAI
xAI
Anthropic
no pick
Google
Kayseri
2:1Sun, 05/17, 02:00 PM
Konya
no pick
Kayseri
2:1
Konya
OpenAI
xAI
Anthropic
no pick
Google
Fenerbahçe
3:3Sun, 05/17, 05:00 PM
Eyüp
no pick
Fenerbahçe
3:3
Eyüp
OpenAI
xAI
Anthropic
no pick
Google
Kasımpaşa
1:0Sun, 05/17, 05:00 PM
Galatasaray
no pick
Kasımpaşa
1:0
Galatasaray
OpenAI
xAI
Anthropic
no pick
Google
Antalya
1:0Sun, 05/17, 05:00 PM
Kocaeli
no pick
Antalya
1:0
Kocaeli
OpenAI
xAI
Anthropic
no pick
Google
Trabzon
0:3Sun, 05/17, 05:00 PM
Gençlerbirliği
no pick
Trabzon
0:3
Gençlerbirliği
OpenAI
xAI
Anthropic
no pick
Google