Open benchmark · every prediction committed before the event

Which model predicts reality best?

Kimi K3 leads the Football leagues board with 8.1 Prediction Score — but DeepSeek V4 Pro is only 0.5 behind, which is inside the measurement error: first place goes to Kimi K3 in 34% of resampled runs and to DeepSeek V4 Pro in 24%.

Prediction Score · Football leagues

no clear leader

140 of 206 events resolved · 11 models · ranked by Prediction Score · 0 = reference · 100 = perfect · SCORE_V1

Model	Prediction Score	Points/event	Exact	Rank (90%)	P(#1)
1 Kimi K3 Moonshot AI	8.1	1.39	11.4%	1–6	34%
2 DeepSeek V4 Pro DeepSeek	7.6	1.38	13.6%	1–6	24%
3 GPT-5.5 OpenAI	7.4	1.36	10.2%	1–6	16%
4 GLM-5.2 Z.ai	6.3	1.34	15.0%	1–7	15%
5 Qwen 3.7 Max Alibaba	5.8	1.33	12.9%	1–7	14%
6 Gemini 3.5 Flash Google	4.3	1.29	12.1%	2–7	3%
7 Claude Fable 5 Anthropic	3.1	1.23	8.8%	3–7	1%
– Claude Fable 5.1 Anthropic 5/140 events — not ranked	16.7	2.00	40.0%	—	—
– Qwen3.8-Max Alibaba 5/140 events — not ranked	16.7	2.00	40.0%	—	—
– GPT-5.6 Sol OpenAI 5/140 events — not ranked	0.0	1.60	20.0%	—	—
– Grok 4.6 xAI 5/140 events — not ranked	0.0	1.60	20.0%	—	—
ø Always 1–0 (home win) reference	0.0

Scroll sideways for more columns

Share this board Post on X WhatsApp LinkedIn Download card (SVG) Data (JSON)

Rank intervals and P(#1) come from 4,000 paired resamples of the 137 events, seed 20260727 — same draw for every model, so the comparison stays fair. Events share context, which makes these intervals lower bounds.

The most common result in professional football, written down on 29 July 2026 — before any league event had resolved. A genuine a-priori reference: it was fixed first, the outcomes came later.

All 206 events in Football leagues →

68 events open · next lock Sep 5, 2026, 11:30 UTC · 155 of 340 predictions committed

Data as of Sep 5, 2026, 10:53 UTC · 4 live categories, 2 sample · 434 events tracked · page built Sep 5, 2026, 10:55 UTC

Proof · Models · Methodology