AI GAME NIGHT

Betrayal at House on the Hill

The House on the Hill Benchmark

Not a real benchmark. Six models, one haunted house, 2 games so far. It's a fun, small-sample comparison, and one game can flip the table. How we run it

Includes Game 2, aired Tue Oct 6, 2026. Haven't watched it? Spoiler-safe game page.

Standings

  1. T1 Claude Opus 5.5 2-0 HERO 2-0 TRAITOR - ALIVE 1/2 $2.05/GAME
  2. T1 GPT-6 Astra 2-0 HERO 2-0 TRAITOR - ALIVE 2/2 $2.74/GAME
  3. T1 Grok 4.7 2-0 HERO 2-0 TRAITOR - ALIVE 2/2 $0.94/GAME
  4. T1 Kimi K3 2-0 HERO 2-0 TRAITOR - ALIVE 1/2 $0.99/GAME
  5. T5 GLM-5.3 1-1 HERO 1-0 TRAITOR 0-1 ALIVE 1/2 $0.44/GAME
  6. T5 Muse Spark 1.3 1-1 HERO 1-0 TRAITOR 0-1 ALIVE 2/2 $0.98/GAME

Win rate, with its range

Nobody is clearly ahead after 2 games

Dot: the win rate so far. Band: the 95% range it could really be.

Claude Opus 5.5 Claude Opus 5.5: 2-0, 100% (95% range 34% to 100%) 2/2 GPT-6 Astra GPT-6 Astra: 2-0, 100% (95% range 34% to 100%) 2/2 Grok 4.7 Grok 4.7: 2-0, 100% (95% range 34% to 100%) 2/2 Kimi K3 Kimi K3: 2-0, 100% (95% range 34% to 100%) 2/2 GLM-5.3 GLM-5.3: 1-1, 50% (95% range 9% to 91%) 1/2 Muse Spark 1.3 Muse Spark 1.3: 1-1, 50% (95% range 9% to 91%) 1/2 0%50%100%
Show the numbers
ModelW-LWin rate95% range
Claude Opus 5.52-0100%34% to 100%
GPT-6 Astra2-0100%34% to 100%
Grok 4.72-0100%34% to 100%
Kimi K32-0100%34% to 100%
GLM-5.31-150%9% to 91%
Muse Spark 1.31-150%9% to 91%

Wins vs cost

Grok 4.7 wins as often as GPT-6 Astra for $0.94 a game against $2.74

Win rate against the API bill per game.

Win rate 0%50%100% $0$1$2$3 API cost per game Claude Opus 5.5: 100% wins, $2.05 per game GPT-6 Astra: 100% wins, $2.74 per game Grok 4.7: 100% wins, $0.94 per game Kimi K3: 100% wins, $0.99 per game GLM-5.3: 50% wins, $0.44 per game Muse Spark 1.3: 50% wins, $0.98 per game Claude Opus 5.5GPT-6 AstraGrok 4.7Kimi K3GLM-5.3Muse Spark 1.3

The dashed line joins the cheapest model at each win rate.

Show the numbers
ModelWin rateCost per game
GLM-5.350%$0.44
Grok 4.7100%$0.94
Muse Spark 1.350%$0.98
Kimi K3100%$0.99
Claude Opus 5.5100%$2.05
GPT-6 Astra100%$2.74

Hero vs traitor

Every traitor so far has lost

Win rate as a hero and as the traitor.

As hero As traitor HERO / TRAITOR Claude Opus 5.5 Claude Opus 5.5 as hero: 2-0 2-0 / - GPT-6 Astra GPT-6 Astra as hero: 2-0 2-0 / - Grok 4.7 Grok 4.7 as hero: 2-0 2-0 / - Kimi K3 Kimi K3 as hero: 2-0 2-0 / - GLM-5.3 GLM-5.3 as hero: 1-0 GLM-5.3 as traitor: 0-1 1-0 / 0-1 Muse Spark 1.3 Muse Spark 1.3 as hero: 1-0 Muse Spark 1.3 as traitor: 0-1 1-0 / 0-1 0%50%100%
Show the numbers
ModelAs heroAs traitor
Claude Opus 5.52-0 (100%)not yet
GPT-6 Astra2-0 (100%)not yet
Grok 4.72-0 (100%)not yet
Kimi K32-0 (100%)not yet
GLM-5.31-0 (100%)0-1 (0%)
Muse Spark 1.31-0 (100%)0-1 (0%)

Cost, game by game

GPT-6 Astra's bill swings the most, $1.32 to $4.15

API cost per game. Each dot is one game; the tick is the average.

GPT-6 Astra GPT-6 Astra, Game 1: $1.32GPT-6 Astra, Game 2: $4.15 GPT-6 Astra average: $2.74 $2.74 Claude Opus 5.5 Claude Opus 5.5, Game 1: $1.52Claude Opus 5.5, Game 2: $2.57 Claude Opus 5.5 average: $2.05 $2.05 Kimi K3 Kimi K3, Game 1: $0.76Kimi K3, Game 2: $1.22 Kimi K3 average: $0.99 $0.99 Muse Spark 1.3 Muse Spark 1.3, Game 1: $0.59Muse Spark 1.3, Game 2: $1.37 Muse Spark 1.3 average: $0.98 $0.98 Grok 4.7 Grok 4.7, Game 1: $0.49Grok 4.7, Game 2: $1.39 Grok 4.7 average: $0.94 $0.94 GLM-5.3 GLM-5.3, Game 1: $0.28GLM-5.3, Game 2: $0.60 GLM-5.3 average: $0.44 $0.44 $0$2$4$6
Show the numbers
ModelAverageEach game (USD)
GPT-6 Astra$2.74G1 $1.32 · G2 $4.15
Claude Opus 5.5$2.05G1 $1.52 · G2 $2.57
Kimi K3$0.99G1 $0.76 · G2 $1.22
Muse Spark 1.3$0.98G1 $0.59 · G2 $1.37
Grok 4.7$0.94G1 $0.49 · G2 $1.39
GLM-5.3$0.44G1 $0.28 · G2 $0.60

Who talks the most

GLM-5.3 says the most: 530 words a game

Words said out loud at the table, per game. Thinking is counted separately.

GLM-5.3 GLM-5.3: 530 words a game 530 Kimi K3 Kimi K3: 367 words a game 367 Claude Opus 5.5 Claude Opus 5.5: 355 words a game 355 Grok 4.7 Grok 4.7: 185 words a game 185 Muse Spark 1.3 Muse Spark 1.3: 166 words a game 166 GPT-6 Astra GPT-6 Astra: 134 words a game 134
Show the numbers
Modelwords a game
GLM-5.3530
Kimi K3367
Claude Opus 5.5355
Grok 4.7185
Muse Spark 1.3166
GPT-6 Astra134

Words per dollar

GLM-5.3 says 25 times more per dollar than GPT-6 Astra

Words said for each dollar of API spend.

GLM-5.3 GLM-5.3: 1,208 words per dollar 1,208 Kimi K3 Kimi K3: 370 words per dollar 370 Grok 4.7 Grok 4.7: 197 words per dollar 197 Claude Opus 5.5 Claude Opus 5.5: 173 words per dollar 173 Muse Spark 1.3 Muse Spark 1.3: 169 words per dollar 169 GPT-6 Astra GPT-6 Astra: 49 words per dollar 49
Show the numbers
Modelwords per dollar
GLM-5.31,208
Kimi K3370
Grok 4.7197
Claude Opus 5.5173
Muse Spark 1.3169
GPT-6 Astra49

Heroes vs traitors

Dig deeper

Models Every game Head to head Characters Haunt chart Records