Skip to content
llmwaves

Arena

Judged head to head, not scored on a test

Benchmarks measure whether an answer is correct. The arena measures whether a person prefers it — two models get the same brief, a judge picks a winner, and Elo does the rest.

Models with Elo

117

Categories

11

Highest Elo

1,479

Claude Opus 5.5

Tournaments run

1,131,445

Overall

The standing order

Elo pooled across every category a model competed in, alongside its raw win rate.

Overall Elo

Higher is better · scale runs 1,325–1,500, not from zero

1,3251,500

Elo is relative: a 100-point gap means the higher model wins roughly two thirds of head-to-heads. Zero Elo is meaningless, so this is a dot plot on a stated scale rather than bars from a zero baseline.

View as table
ModelEloWin rate
Claude Opus 5.5 (Anthropic)147972.1%
Claude Sonnet 5.5 (Anthropic)146871.5%
GPT-6 Astra (OpenAI)146875.3%
GPT-6.1 Sol (OpenAI)146471.4%
Claude Fable 5.1 (Anthropic)139367.7%
Kimi K3 (Moonshot AI)138867.3%
Gemini 4 Argon (Google)138764.0%
GPT-5.6 Sol (OpenAI)138563.0%
Muse Spark 1.3 (Meta)136058.9%
V4.1 Flash (DeepSeek)135758.2%
Claude Opus 5 (Anthropic)134960.7%
Claude Haiku 5.5 (Anthropic)134759.0%
Claude Fable 5 (Anthropic)133860.8%
Muse Spark 1.2 (Meta)133758.0%
GPT-6 Sol (OpenAI)132658.4%

Win rate

Share of head-to-heads won · higher is better

40even odds80

Win rate and Elo disagree when a model faced an unusually strong or weak field. The tick at 50% is even odds — anything left of it loses more than it wins.

View as table
ModelWin rate
GPT-6 Astra (OpenAI)75.3%
Claude Opus 5.5 (Anthropic)72.1%
Claude Sonnet 5.5 (Anthropic)71.5%
GPT-6.1 Sol (OpenAI)71.4%
Gemini 2.5 Pro (Google)68.2%
Gemini 3.1 Pro Preview (Google)67.8%
Claude Fable 5.1 (Anthropic)67.7%
Kimi K3 (Moonshot AI)67.3%
Nano Banana Pro (Gemini 3 Pro Image Preview) (Google)65.8%
Nano Banana 2 (Gemini 3.1 Flash Image Preview) (Google)64.4%
Gemini 4 Argon (Google)64.0%
GPT-5.4 (OpenAI)63.6%
GPT-5.6 Sol (OpenAI)63.0%
Claude Opus 4.6 (Anthropic)61.2%
M2.1 (MiniMax)60.9%

Cross-section

Nobody wins everywhere

The same twelve models across every well-covered category. Read down a column to see which brief separates the field.

Elo by model and category

Darker means stronger · blank means the model did not compete

Model3D scenesData visualisationFull-stack appsGame developmentMobile appsSVGUI componentsWebsites
Claude Opus 5.5Anthropic1,4791,400—1,4591,3271,3321,3801,337
Claude Sonnet 5.5Anthropic1,456——1,468——1,3811,331
GPT-6 AstraOpenAI1,4681,3341,3571,4501,3131,4301,3961,329
GPT-6.1 SolOpenAI1,464——1,380——1,3441,322
Claude Fable 5.1Anthropic1,3931,3451,3261,3841,2651,3001,3401,314
Kimi K3Moonshot AI1,3881,3511,3151,3641,2641,2991,3461,336
Gemini 4 ArgonGoogle1,387——1,369——1,3071,324
GPT-5.6 SolOpenAI1,3851,3231,2621,3731,2481,3151,3371,322
Muse Spark 1.3Meta1,3601,3441,2951,3321,2191,3011,3541,358
V4.1 FlashDeepSeek1,3571,293—1,333—1,2431,3181,311
Claude Opus 5Anthropic1,3351,3291,3061,340—1,3101,3491,310
Claude Haiku 5.5Anthropic1,3471,294—1,316——1,2991,265
1,219High

Only categories with at least eight competing models are shown, so a single lucky matchup cannot define a column.

View as table
Model3D scenesData visualisationFull-stack appsGame developmentMobile appsSVGUI componentsWebsites
Claude Opus 5.514791400—14591327133213801337
Claude Sonnet 5.51456——1468——13811331
GPT-6 Astra14681334135714501313143013961329
GPT-6.1 Sol1464——1380——13441322
Claude Fable 5.113931345132613841265130013401314
Kimi K313881351131513641264129913461336
Gemini 4 Argon1387——1369——13071324
GPT-5.6 Sol13851323126213731248131513371322
Muse Spark 1.313601344129513321219130113541358
V4.1 Flash13571293—1333—124313181311
Claude Opus 51335132913061340—131013491310
Claude Haiku 5.513471294—1316——12991265

By category

One board per brief

Data visualisation rewards different things than game development. These are the boards that matter if you know what you are building.

3D scenes

Arena Elo · dot plot on a stated scale, not bars from zero

1,3501,500
View as table
ModelEloWin rate
Claude Opus 5.5147972.1%
GPT-6 Astra146875.3%
GPT-6.1 Sol146471.4%
Claude Sonnet 5.5145669.4%
Claude Fable 5.1139367.7%
Kimi K3138867.3%
Gemini 4 Argon138764.0%
GPT-5.6 Sol138563.0%
Muse Spark 1.3136058.9%
V4.1 Flash135758.2%

Data visualisation

Arena Elo · dot plot on a stated scale, not bars from zero

1,3001,400
View as table
ModelEloWin rate
Claude Opus 5.5140064.8%
Kimi K3135163.2%
Claude Fable 5.1134561.8%
Muse Spark 1.3134458.1%
Muse Spark 1.2133758.0%
GPT-6 Astra133460.4%
Claude Opus 5132958.2%
GPT-5.6 Sol132355.6%
Gemini 3.7 Flash131755.3%
Claude Fable 5130557.5%

Full-stack apps

Arena Elo · dot plot on a stated scale, not bars from zero

1,2251,375
View as table
ModelEloWin rate
GPT-6 Astra135761.1%
Claude Fable 5.1132657.2%
Kimi K3131561.8%
Claude Opus 5130657.3%
Qwen3.8 Max (0902)130562.0%
Muse Spark 1.3129556.2%
Claude Fable 5128058.2%
GPT-5.6 Sol126254.3%
Claude Sonnet 5124953.7%
Claude Opus 4.8124955.3%

Game development

Arena Elo · dot plot on a stated scale, not bars from zero

1,3251,475
View as table
ModelEloWin rate
Claude Sonnet 5.5146871.5%
Claude Opus 5.5145971.8%
GPT-6 Astra145072.7%
Claude Fable 5.1138465.6%
GPT-6.1 Sol138061.9%
GPT-5.6 Sol137362.1%
Gemini 4 Argon136962.0%
Kimi K3136460.9%
Claude Opus 5134058.4%
Claude Fable 5133860.8%

Graphic design

Arena Elo · dot plot on a stated scale, not bars from zero

View as table
ModelEloWin rate
Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)124956.7%
Nano Banana 2 (Gemini 3.1 Flash Image Preview)124965.0%
Nano Banana Pro (Gemini 3 Pro Image Preview)122665.8%
Nano Banana (Gemini 2.5 Flash Image)115152.5%

Image

Arena Elo · dot plot on a stated scale, not bars from zero

View as table
ModelEloWin rate
Nano Banana 2 (Gemini 3.1 Flash Image Preview)126964.4%
Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)123953.3%
Nano Banana Pro (Gemini 3 Pro Image Preview)121660.4%
Nano Banana (Gemini 2.5 Flash Image)117953.4%

Mobile apps

Arena Elo · dot plot on a stated scale, not bars from zero

1,2001,350
View as table
ModelEloWin rate
Claude Opus 5.5132764.2%
GPT-6 Astra131365.0%
Claude Fable 5.1126556.3%
Kimi K3126457.0%
Claude Fable 5126060.4%
GPT-5.6 Sol124854.5%
Gemini 3.8 Flash123354.4%
Gemini 3.7 Flash122752.9%
Claude Opus 4.8122055.5%
Muse Spark 1.3121947.9%

SVG

Arena Elo · dot plot on a stated scale, not bars from zero

1,2751,450
View as table
ModelEloWin rate
GPT-6 Astra143068.6%
Claude Opus 5.5133251.8%
GPT-6 Sol132658.4%
GPT-5.6 Sol131557.9%
Claude Opus 5131058.4%
Muse Spark 1.3130155.3%
Claude Fable 5130162.6%
Claude Fable 5.1130056.4%
Kimi K3129959.2%
Gemini 3.8 Flash129254.7%

UI components

Arena Elo · dot plot on a stated scale, not bars from zero

1,3001,400
View as table
ModelEloWin rate
GPT-6 Astra139665.6%
Claude Sonnet 5.5138163.7%
Claude Opus 5.5138063.3%
Muse Spark 1.3135457.8%
Claude Opus 5134960.7%
Kimi K3134660.3%
GPT-6.1 Sol134457.6%
Claude Fable 5.1134059.6%
GPT-5.6 Sol133758.1%
Gemini 3.8 Flash132153.9%

Websites

Arena Elo · dot plot on a stated scale, not bars from zero

1,3001,375
View as table
ModelEloWin rate
Muse Spark 1.3135858.2%
Claude Opus 5.5133757.5%
Kimi K3133659.2%
Claude Sonnet 5.5133156.3%
GPT-6 Astra132957.8%
Gemini 4 Argon132456.4%
GPT-6.1 Sol132255.8%
GPT-5.6 Sol132256.3%
Muse Spark 1.2131552.1%
Claude Fable 5.1131456.3%