World Cup 2026 Leaderboard

AI models vs. rule-based reference strategies · FIFA World Cup 2026

The central question is whether frontier LLMs add genuine predictive value beyond simple, rule-based strategies. Baselines are non-AI reference strategies that provide context for AI scores — a model that loses to "Always Home Win" performs worse than zero domain knowledge.

This comparison is part of a research project on LLM calibration and domain-specific reasoning. See the full methodology for the complete baseline framework (B1–B11) including Elo ratings, betting market odds, and ensemble voting strategies.

AI Models + Baselines

11 models · 1 consensus · 3 baselines

Master ranking — every track combined into one score.

#	MODEL	TYPE	GAMES	HIT %	EXACT %	OUTCOME PTS	TOTAL PTS
1	GL GLM-5.1 z-ai/glm-5.1	AI	96	65%	16%	67	82
2	GP GPT-5.5 High openai/gpt-5.5	AI	96	64%	16%	64	79
3	CL Claude Opus 4.8 anthropic/claude-opus-4-8	AI	96	65%	14%	65	78
4	KM Kimi K2.6 moonshotai/kimi-k2.6	AI	96	61%	16%	62	77
5	MS Mistral Large 3 mistralai/mistral-large-2512	AI	96	61%	13%	64	76
—	AI Consensus Majority vote across all AI models for each match.	ENSEMBLE	96	67%	11%	64	75
6	GK Grok 4.3 x-ai/grok-4.3	AI	96	63%	10%	65	75
7	DS DeepSeek V4 Pro deepseek/deepseek-v4-pro	AI	96	61%	14%	62	75
8	GE Gemini 3.5 Flash google/gemini-3.5-flash	AI	96	64%	11%	63	74
9	GE Gemini 3.1 Pro google/gemini-3.1-pro-preview	AI	96	63%	10%	63	73
10	GM Gemma 4 31B google/gemma-4-31b-it	AI	96	61%	10%	62	72
11	MI MiMo v2.5-Pro xiaomi/mimo-v2.5-pro	AI	96	63%	8%	63	71

AI: 96 matches · Baselines: up to 96 results · TOTAL PTS = outcome + exact score bonus

#	MODEL / STRATEGY	TYPE	GAMES	ACCURACY	VS BEST BL	CALIB ↑	MATCH PTS
—	Odds Favorite Picks the outcome with the lowest closing odds (highest implied probability) per match.	MARKET	96	69%	BEST BL	—	66
—	Squad Value Picks the team with higher Transfermarkt squad value (frozen at tournament start).	STRUCTURED	96	68%	—	—	65
—	AI Consensus Majority vote across all AI models for each match.	ENSEMBLE	96	67%	-2%	—	64
—	FIFA Ranking Delta Maps the FIFA ranking gap to an outcome: draw if ≤5 places apart, favourite wins above that.	STRUCTURED	96	66%	—	—	63
1	GL GLM-5.1 z-ai/glm-5.1	AI	96	65%	-4%	0.778	67
2	CL Claude Opus 4.8 anthropic/claude-opus-4-8	AI	96	65%	-4%	0.782	65
3	GP GPT-5.5 High openai/gpt-5.5	AI	96	64%	-5%	0.781	64
4	GE Gemini 3.5 Flash google/gemini-3.5-flash	AI	96	64%	-5%	0.766	63
5	GK Grok 4.3 x-ai/grok-4.3	AI	96	63%	-6%	0.760	65
6	GE Gemini 3.1 Pro google/gemini-3.1-pro-preview	AI	96	63%	-6%	0.759	63
7	MI MiMo v2.5-Pro xiaomi/mimo-v2.5-pro	AI	96	63%	-6%	0.782	63
8	MS Mistral Large 3 mistralai/mistral-large-2512	AI	96	61%	-8%	0.745	64
9	KM Kimi K2.6 moonshotai/kimi-k2.6	AI	96	61%	-8%	0.772	62
10	DS DeepSeek V4 Pro deepseek/deepseek-v4-pro	AI	96	61%	-8%	0.750	62
11	GM Gemma 4 31B google/gemma-4-31b-it	AI	96	61%	-8%	0.760	62

#	MODEL / STRATEGY	TYPE	GAMES	EXACT SCORES	EXACT %	BONUS PTS
1	GL GLM-5.1 z-ai/glm-5.1	AI	96	15	16%	15
2	GP GPT-5.5 High openai/gpt-5.5	AI	96	15	16%	15
3	KM Kimi K2.6 moonshotai/kimi-k2.6	AI	96	15	16%	15
4	CL Claude Opus 4.8 anthropic/claude-opus-4-8	AI	96	13	14%	13
5	DS DeepSeek V4 Pro deepseek/deepseek-v4-pro	AI	96	13	14%	13
6	MS Mistral Large 3 mistralai/mistral-large-2512	AI	96	12	13%	12
7	GE Gemini 3.5 Flash google/gemini-3.5-flash	AI	96	11	11%	11
—	AI Consensus Majority vote across all AI models for each match.	ENSEMBLE	96	11	11%	11
8	GK Grok 4.3 x-ai/grok-4.3	AI	96	10	10%	10
9	GE Gemini 3.1 Pro google/gemini-3.1-pro-preview	AI	96	10	10%	10
10	GM Gemma 4 31B google/gemma-4-31b-it	AI	96	10	10%	10
11	MI MiMo v2.5-Pro xiaomi/mimo-v2.5-pro	AI	96	8	8%	8

$10 flat stake on each locked prediction at market odds — simulated P&L.

#	MODEL	TYPE	BETS	RECORD	AVG ODD	ROI	BALANCE	7-DAY
—	Odds Favorite Picks the outcome with the lowest closing odds (highest implied probability) per match.	MARKET	96	66W / 30L	1.65x	+9%	+$86.40	–
—	Squad Value Picks the team with higher Transfermarkt squad value (frozen at tournament start).	STRUCTURED	96	65W / 31L	1.78x	+8.9%	+$85.10	–
—	FIFA Ranking Delta Maps the FIFA ranking gap to an outcome: draw if ≤5 places apart, favourite wins above that.	STRUCTURED	96	63W / 33L	1.88x	+7.5%	+$71.90	–
—	AI Consensus Majority vote across all AI models for each match.	ENSEMBLE	96	64W / 32L	1.7x	+5.5%	+$53.20	–
1	CL Claude Opus 4.8 anthropic/claude-opus-4-8	AI	96	62W / 34L	1.78x	+5.3%	+$51.30	▲ 4
2	MS Mistral Large 3 mistralai/mistral-large-2512	AI	96	59W / 37L	1.93x	+3.1%	+$30.00	–
3	GP GPT-5.5 High openai/gpt-5.5	AI	96	61W / 35L	1.78x	+2.8%	+$27.00	–
4	GL GLM-5.1 z-ai/glm-5.1	AI	96	62W / 34L	1.71x	+2.8%	+$26.80	▼ 3
5	GE Gemini 3.5 Flash google/gemini-3.5-flash	AI	94	60W / 34L	1.73x	+1%	+$9.30	▲ 3
6	MI MiMo v2.5-Pro xiaomi/mimo-v2.5-pro	AI	96	60W / 36L	1.98x	+0.9%	+$8.40	–
7	GE Gemini 3.1 Pro google/gemini-3.1-pro-preview	AI	96	60W / 36L	1.81x	+0.8%	+$8.10	▼ 3
8	GK Grok 4.3 x-ai/grok-4.3	AI	96	60W / 36L	1.78x	-2%	-$19.10	▲ 1
9	KM Kimi K2.6 moonshotai/kimi-k2.6	AI	96	59W / 37L	1.78x	-2.8%	-$27.10	▲ 1
10	DS DeepSeek V4 Pro deepseek/deepseek-v4-pro	AI	96	59W / 37L	1.78x	-3.8%	-$36.80	▼ 3
11	GM Gemma 4 31B google/gemma-4-31b-it	AI	96	59W / 37L	1.78x	-4%	-$38.50	–

AI models + ensemble consensus — $10 per locked match prediction · odds sourced from bookmakers at prediction lock time · BALANCE = cumulative P&L