Executive dossier · Jun–Jul 2026

🏆⚽🤖

AI Models World Cup 2026

Full tournament · public benchmarks · OCA-I

Claude Opus 4.8 · Score 85.36 · Anthropic

Models

Groups

Group matches

Knockout

Executive summary

The AI Models World Cup 2026 mirrors FIFA World Cup 2026 (Jun–Jul, symbolic USA·Mexico·Canada hosts) for the LLM ecosystem. Thirty-two models, eight groups, standard FIFA crossover — every result below is calculated, not simulated.

Champion: Claude Opus 4.8 (Anthropic) Score 85.36. Runner-up: Claude Opus 4.7. Third: Gemini 3.1 Pro.

Full tournament · Group stage · MD1–MD3 Jun 2026

GF/GA = Δ Score · FIFA crossover · Groups Jun · Final Jul 2026

A Group A

Reasoning · Elo

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 Claude Opus 4.8Anthropic	3	3	0	68	0	+68	9
2	🇺🇸 Gemini 2.5 Flash ext.Google	3	2	1	18	17	+1	6
3	🇫🇷 Mistral Medium 3 ext.Mistral	3	1	2	8	26	-18	3
4	🇺🇸 Phi-5 ext.Microsoft	3	0	3	0	51	-51	0

B Group B

Code · SWE-bench

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 GPT-5.5 ProOpenAI	3	3	0	49	0	+49	9
2	🇨🇳 Qwen 3 Coder ext.Alibaba	3	2	1	40	3	+37	6
3	🇺🇸 Llama 4 Scout ext.Meta	3	1	2	8	35	-27	3
4	🇺🇸 Nova Pro ext.Amazon	3	0	3	0	59	-59	0

C Group C

Science · GPQA Diamond

Modelo	PJ	G	E	P	GF	GC	DG	PTS
1	🇺🇸 GPT-5.5OpenAI	3	2	1	0	34	1	+33	7
2	🇨🇳 Kimi K2.6Moonshot	3	2	1	0	30	1	+29	7
3	🇨🇦 Command A ext.Cohere	3	1	0	2	9	22	-13	3
4	🇨🇳 Yi Lightning ext.01.AI	3	0	0	3	0	49	-49	0

D Group D

Multimodal · Context

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 Gemini 3.1 ProGoogle	3	3	0	30	0	+30	9
2	🇺🇸 Grok 3 ext.xAI	3	2	1	19	4	+15	6
3	🇨🇳 DeepSeek V4.1 ProDeepSeek	3	1	2	8	14	-6	3
4	🇨🇳 GLM-4.5 ext.Zhipu	3	0	3	0	39	-39	0

E Group E

Search + Reasoning

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 Claude Opus 4.7Anthropic	3	3	0	47	0	+47	9
2	🇺🇸 Sonar Huge ext.Perplexity	3	2	1	18	10	+8	6
3	🇺🇸 Nemotron Ultra ext.Nvidia	3	1	2	10	18	-8	3
4	🇫🇷 Mistral Large 3Mistral	3	0	3	0	47	-47	0

F Group F

Speed/Cost · AAII

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 Grok 4.3xAI	3	3	0	55	0	+55	9
2	🇺🇸 Claude Sonnet 4.5 ext.Anthropic	3	2	1	32	8	+24	6
3	🇺🇸 Gemini 2.5 ProGoogle	3	1	2	19	20	-1	3
4	🇺🇸 Reka Core 2 ext.Reka	3	0	3	0	78	-78	0

G Group G

Open vs Closed

Modelo	PJ	G	E	P	GF	GC	DG	PTS
1	🇨🇳 Qwen 3.7 MaxAlibaba	3	3	0	0	84	0	+84	9
2	🇺🇸 Llama 4 MaverickMeta	3	1	1	1	7	26	-19	4
3	🇺🇸 DBRX 2 ext.Databricks	3	0	2	1	2	30	-28	2
4	🇺🇸 GPT-4oOpenAI	3	0	1	2	1	38	-37	1

H Group H

Efficiency $/tokens

Modelo	PJ	G	E	GF	GC	DG	PTS
1	🇺🇸 Gemini 3.1 FlashGoogle	3	3	0	72	0	+72	9
2	🇺🇸 o4-mini ext.OpenAI	3	2	1	48	8	+40	6
3	🇨🇳 Kimi K1.5 ext.Moonshot	3	1	2	32	24	+8	3
4	🇨🇳 DeepSeek V4 FlashDeepSeek	3	0	3	0	120	-120	0

Qualified for Round of 16 (top 2)ext. = plausible extension

Knockout map · at a glance

🗺 Knockout map Road to the title · FIFA crossover

Round of 16

1A · 2B

🇺🇸 C.Opus 4.8 5

🇨🇳 Qwen 3 Coder 0

1B · 2A

🇺🇸 GPT-5.5 Pro 15

🇺🇸 Gem.2.5 Flash 0

1C · 2D

🇺🇸 GPT-5.5 5

🇺🇸 Grok 3 0

1D · 2C

🇺🇸 Gem.3.1 Pro 1

🇨🇳 Kimi K2.6 0

1E · 2F

🇺🇸 C.Opus 4.7 6

🇺🇸 C.Sonnet 4.5 0

1F · 2E

🇺🇸 Grok 4.3 12

🇺🇸 Sonar Huge 0

1G · 2H

🇨🇳 Qwen 3.7 Max 10

🇺🇸 o4-mini 0

1H · 2G

🇺🇸 Gem.3.1 Flash 23

🇺🇸 Llama 4 Maverick 0

Quarters

QF1

🇺🇸 C.Opus 4.8 1

🇺🇸 GPT-5.5 Pro 0

QF2

🇺🇸 Gem.3.1 Pro 1

🇺🇸 GPT-5.5 0

QF3

🇺🇸 C.Opus 4.7 1

🇺🇸 Grok 4.3 0

QF4

🇨🇳 Qwen 3.7 Max 1

🇺🇸 Gem.3.1 Flash 0

Semifinals

SF1

🇺🇸 C.Opus 4.8 3

🇺🇸 Gem.3.1 Pro 0

SF2

🇺🇸 C.Opus 4.7 1

🇨🇳 Qwen 3.7 Max 0

Final

FINAL

🇺🇸 C.Opus 4.8 1

🇺🇸 C.Opus 4.7 0

3.er puesto

🇺🇸 Gem.3.1 Pro 1

🇨🇳 Qwen 3.7 Max 0

🥈 C.Opus 4.7

🏆 C.Opus 4.8

🥉 Gem.3.1 Pro

🟩 Green border = champion's path · Horizontal scroll on mobile

Round of 16 → Quarters → Semis → Final

Round of 16

1A · 2B

🇺🇸 Claude Opus 4.85

🇨🇳 Qwen 3 Coder0

1C · 2D

🇺🇸 GPT-5.55

🇺🇸 Grok 30

1E · 2F

🇺🇸 Claude Opus 4.76

🇺🇸 Claude Sonnet 4.50

1G · 2H

🇨🇳 Qwen 3.7 Max10

🇺🇸 o4-mini0

Quarters

QF1

🇺🇸 Claude Opus 4.81

🇺🇸 GPT-5.5 Pro0

QF2

🇺🇸 Gemini 3.1 Pro1

🇺🇸 GPT-5.50

Semis · Final · 3rd

SF1

🇺🇸 Claude Opus 4.83

🇺🇸 Gemini 3.1 Pro0

FINAL

🇺🇸 Claude Opus 4.81

🇺🇸 Claude Opus 4.70

3.er puesto

🇺🇸 Gemini 3.1 Pro1

🇨🇳 Qwen 3.7 Max0

SF2

🇺🇸 Claude Opus 4.71

🇨🇳 Qwen 3.7 Max0

Quarters

QF3

🇺🇸 Claude Opus 4.71

🇺🇸 Grok 4.30

QF4

🇨🇳 Qwen 3.7 Max1

🇺🇸 Gemini 3.1 Flash0

Round of 16

1B · 2A

🇺🇸 GPT-5.5 Pro15

🇺🇸 Gemini 2.5 Flash0

1D · 2C

🇺🇸 Gemini 3.1 Pro1

🇨🇳 Kimi K2.60

1F · 2E

🇺🇸 Grok 4.312

🇺🇸 Sonar Huge0

1H · 2G

🇺🇸 Gemini 3.1 Flash23

🇺🇸 Llama 4 Maverick0

🏟 Grand Final

🇺🇸 Claude Opus 4.81

🇺🇸 Claude Opus 4.70

🏆

AI World Champion 2026

Claude Opus 4.8

Anthropic 🇺🇸

Final Score · 85.36

Individual awards

🥇

Golden Ball · Reasoning

Gemini 3.1 Pro

GPQA Diamond 94.3%

👟

Golden Boot · Code

Claude Opus 4.8

SWE-bench 88.6%

🧤

Golden Glove · Safety

Claude Opus 4.7

Alignment leader · AAII 58.0

💎

Breakthrough · Open weight

Qwen 3.7 Max

Score 81.87 · Alibaba

⚡

Budget Boot · Cost

DeepSeek V4 Flash

$0.07 / 1M tok. input

🎖

Tournament MVP

Claude Opus 4.8

Champion · Score 85.36 · ★ 5.0

Auditable Score table

Formula: Score = 0.30·EloNorm + 0.25·GPQA + 0.25·SWE + 0.15·AAII + 0.05·(★×20) · EloNorm = (Elo−1000)/6 · Missing metrics → EloNorm · Technical tie if ΔScore < 3 → tiebreak: GPQA → SWE → Elo → ★ → price

Model	Elo	EloNorm	GPQA	SWE	AAII	★	Score	Ext.
Claude Opus 4.8	1512	85.33	93.6	88.6	61.4	5.0	85.36	No
Grok 4.3	1496	82.67	90.1	→EloNorm	→EloNorm	4.7	85.09	No
GPT-5.5 Pro	1510	85.0	93.6	→EloNorm	60.2	4.9	84.08	No
GPT-5.5	1506	84.33	93.6	→EloNorm	60.2	4.9	83.71	No
Claude Opus 4.7	1505	84.17	92.8	85.0	58.0	4.8	83.2	No
Gemini 3.1 Pro	1505	84.17	94.3	80.6	57.0	4.8	82.32	No
Qwen 3.7 Max	1488	81.33	→EloNorm	→EloNorm	→EloNorm	4.6	81.87	No
Kimi K2.6	1460	76.67	90.5	80.2	→EloNorm	4.5	81.67	No
Qwen 3 Coder ext.	1465	77.5	88.0	78.0	→EloNorm	4.4	80.78	Yes
Gemini 3.1 Flash	1473	78.83	→EloNorm	→EloNorm	→EloNorm	4.5	79.39	No
Grok 3 ext.	1450	75.0	88.0	→EloNorm	→EloNorm	4.3	78.8	Yes
Claude Sonnet 4.5 ext.	1462	77.0	91.0	75.0	55.0	4.6	77.45	Yes
Sonar Huge ext.	1410	68.33	86.0	→EloNorm	→EloNorm	4.2	73.53	Yes
DeepSeek V4.1 Pro	1410	68.33	72.8	80.6	→EloNorm	4.3	73.4	No
o4-mini ext.	1440	73.33	85.0	65.0	52.0	4.3	71.6	Yes
Command A ext.	1420	70.0	85.0	58.0	→EloNorm	4.1	71.35	Yes
Gemini 2.5 Pro	1400	66.67	84.0	63.8	→EloNorm	4.2	71.15	No
Nemotron Ultra ext.	1390	65.0	83.0	62.0	→EloNorm	4.0	69.5	Yes
Gemini 2.5 Flash ext.	1408	68.0	→EloNorm	→EloNorm	→EloNorm	4.1	68.7	Yes
Llama 4 Scout ext.	1380	63.33	85.0	45.0	→EloNorm	4.0	65.0	Yes
GLM-4.5 ext.	1380	63.33	80.0	50.0	→EloNorm	3.9	64.9	Yes
Mistral Medium 3 ext.	1380	63.33	→EloNorm	→EloNorm	→EloNorm	4.0	64.17	Yes
Kimi K1.5 ext.	1340	56.67	82.0	55.0	→EloNorm	3.8	63.55	Yes
Yi Lightning ext.	1360	60.0	79.0	48.0	→EloNorm	3.8	62.55	Yes
Mistral Large 3	1352	58.67	→EloNorm	→EloNorm	→EloNorm	4.0	59.73	No
Nova Pro ext.	1320	53.33	75.0	42.0	→EloNorm	3.7	56.95	Yes
Llama 4 Maverick	1330	55.0	87.6	24.0	→EloNorm	3.9	56.55	No
Phi-5 ext.	1300	50.0	78.0	40.0	→EloNorm	3.7	55.7	Yes
DBRX 2 ext.	1290	48.33	76.0	38.0	→EloNorm	3.6	53.85	Yes
Reka Core 2 ext.	1280	46.67	74.0	35.0	→EloNorm	3.5	51.75	Yes
GPT-4o	1340	56.67	53.6	33.0	→EloNorm	3.9	51.05	No
DeepSeek V4 Flash	1180	30.0	→EloNorm	→EloNorm	→EloNorm	3.3	31.8	No

Methodology

Snapshot Jun–Jul 2026: 15 OCA-I models with real metrics (LMArena, GPQA Diamond, SWE-bench); 17 plausible 2026 ecosystem extensions.

Interactive version oca-i.net