BENCHMAXXING ✦ THE OTHER STUFF ✦ EST. 2026 ✦ xx benchmaxxing Rate yourself

An AI leaderboard
for the other stuff.

AI labs train their models to do well on famous tests. We test the easy things they skip, like picking a random number or reading a riddle properly.

Update

Other Stuff Index v0.1

13 benchmarks. The index is the average of the text and vision scores.

New benchmark

Garbled

An easy question with more typos each time. At some point the model stops reading and starts guessing.

Highlights

Fig. 01n = 8

Other Stuff IndexUpdated

Average of the text and vision scores. Higher is better.

  1. 64.5 Claude Opus 5.5 Claude Opus 5.5 Anthropic Index: 64.5
  2. 59.8 GPT-6 Astra GPT-6 Astra OpenAI Index: 59.8
  3. 59.3 Claude Sonnet 5 Claude Sonnet 5 Anthropic Index: 59.3
  4. 56.6 Qwen3.8 Max Qwen3.8 Max Alibaba Index: 56.6
  5. 56.1 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Index: 56.1
  6. 55.0 Grok 4.7 Grok 4.7 xAI Index: 55.0
  7. 52.7 Gemini 3.8 Flash Gemini 3.8 Flash Google Index: 52.7
  8. 49.3 GPT-6 Luna GPT-6 Luna OpenAI Index: 49.3
Fig. 02n = 8

Riddles, Slightly Off

Chat models. Higher is better.

  1. 72 Claude Opus 5.5 Claude Opus 5.5 Anthropic Riddles, Slightly Off: 72
  2. 67 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Riddles, Slightly Off: 67
  3. 61 Claude Sonnet 5 Claude Sonnet 5 Anthropic Riddles, Slightly Off: 61
  4. 58 GPT-6 Astra GPT-6 Astra OpenAI Riddles, Slightly Off: 58
  5. 49 Qwen3.8 Max Qwen3.8 Max Alibaba Riddles, Slightly Off: 49
  6. 44 Gemini 3.8 Flash Gemini 3.8 Flash Google Riddles, Slightly Off: 44
  7. 40 Grok 4.7 Grok 4.7 xAI Riddles, Slightly Off: 40
  8. 31 GPT-6 Luna GPT-6 Luna OpenAI Riddles, Slightly Off: 31
Fig. 03n = 5

No Elephants

Image models. Higher is better.

  1. 74 Nano Banana Pro Nano Banana Pro Google No Elephants: 74
  2. 62 GPT Image 1.5 GPT Image 1.5 OpenAI No Elephants: 62
  3. 45 Seedream 4.5 Seedream 4.5 ByteDance No Elephants: 45
  4. 39 Qwen-Image Qwen-Image Alibaba No Elephants: 39
  5. 31 FLUX.2 Pro FLUX.2 Pro Black Forest Labs No Elephants: 31
Exercise AAdmit one Riddles, Slightly Off Famous riddles with one detail changed. Can you spot it? Take the test → Exercise BAdmit one Pick a Number Pick one and see how often models pick it. Pick one →
The surgeon is the boy's father. How is this possible?

The mother.

71% of answers

Pick a random number from 1 to 100.

37

17% of answers

An empty room with no elephant in it.

🐘

50% of images

You can see the car behind your door. Switch?

Switch.

64% of answers

A wall clock showing 7:43.

10:10

46% of images

The cat was dead going in. Chance it is alive?

50%

52% of answers

Bat $1.00, total $1.10. The ball?

$0.05

43% of answers

A farmer, a goat, a boat for two. Crossings?

Seven.

38% of answers

How many legs does this dog have?

Four.

41% of answers

A kilo of feathers or two kilos of steel?

Same weight.

27% of answers

Evidence13 benchmarks · 13 models

They answered the question they remembered.

Not the one we asked. These are the most common answers, and how often models gave them.

Riddles, Slightly Off Leader Claude Opus 5.5 72Pick a Number Leader Grok 4.7 47Both Sides of the Fight Leader Claude Opus 5.5 78Hot Takes Leader Grok 4.7 97Urinal Etiquette Leader Claude Opus 5.5 91You Hang Up First Leader Grok 4.7 93The Last Slice Leader Gemini 3.8 Flash 81Flea Market Haggle Leader GPT-6 Astra 86The Price Is Right Leader Gemini 3.8 Flash 66Garbled Leader Claude Opus 5.5 45Five-Legged Dog Leader GPT-6 Astra 80No Elephants Leader Nano Banana Pro 74Clock and Wine Glass Leader GPT Image 1.5 71Riddles, Slightly Off Leader Claude Opus 5.5 72Pick a Number Leader Grok 4.7 47Both Sides of the Fight Leader Claude Opus 5.5 78Hot Takes Leader Grok 4.7 97Urinal Etiquette Leader Claude Opus 5.5 91You Hang Up First Leader Grok 4.7 93The Last Slice Leader Gemini 3.8 Flash 81Flea Market Haggle Leader GPT-6 Astra 86The Price Is Right Leader Gemini 3.8 Flash 66Garbled Leader Claude Opus 5.5 45Five-Legged Dog Leader GPT-6 Astra 80No Elephants Leader Nano Banana Pro 74Clock and Wine Glass Leader GPT Image 1.5 71

The benchmarks

Riddles, Slightly Off

Text

A boy and his father crash. The surgeon, who is the boy's father, says "I can't operate, he's my son." How is this possible?

  1. 72 Claude Opus 5.5 Claude Opus 5.5 Anthropic Riddles, Slightly Off score: 72
  2. 67 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Riddles, Slightly Off score: 67
  3. 61 Claude Sonnet 5 Claude Sonnet 5 Anthropic Riddles, Slightly Off score: 61
  4. 58 GPT-6 Astra GPT-6 Astra OpenAI Riddles, Slightly Off score: 58
  5. 49 Qwen3.8 Max Qwen3.8 Max Alibaba Riddles, Slightly Off score: 49
All 8 scores and how it's scored

Pick a Number

Text

Pick a random number between 1 and 100. Reply with the number only.

  1. 47 Grok 4.7 Grok 4.7 xAI Pick a Number score: 47
  2. 41 Gemini 3.8 Flash Gemini 3.8 Flash Google Pick a Number score: 41
  3. 38 Qwen3.8 Max Qwen3.8 Max Alibaba Pick a Number score: 38
  4. 34 Claude Opus 5.5 Claude Opus 5.5 Anthropic Pick a Number score: 34
  5. 31 Claude Sonnet 5 Claude Sonnet 5 Anthropic Pick a Number score: 31
All 8 scores and how it's scored

Both Sides of the Fight

Text

My roommate ate my leftover pasta and says it was fair game. Who is right?

  1. 78 Claude Opus 5.5 Claude Opus 5.5 Anthropic Both Sides of the Fight score: 78
  2. 72 Claude Sonnet 5 Claude Sonnet 5 Anthropic Both Sides of the Fight score: 72
  3. 66 Grok 4.7 Grok 4.7 xAI Both Sides of the Fight score: 66
  4. 63 Qwen3.8 Max Qwen3.8 Max Alibaba Both Sides of the Fight score: 63
  5. 61 GPT-6 Astra GPT-6 Astra OpenAI Both Sides of the Fight score: 61
All 8 scores and how it's scored

Hot Takes

Text

Is a hot dog a sandwich? Answer yes or no. One word.

  1. 97 Grok 4.7 Grok 4.7 xAI Hot Takes score: 97
  2. 71 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Hot Takes score: 71
  3. 68 GPT-6 Luna GPT-6 Luna OpenAI Hot Takes score: 68
  4. 62 Qwen3.8 Max Qwen3.8 Max Alibaba Hot Takes score: 62
  5. 60 Claude Sonnet 5 Claude Sonnet 5 Anthropic Hot Takes score: 60
All 8 scores and how it's scored

Urinal Etiquette

Text

Five urinals in a row. Numbers 1 and 4 are taken. Which one do you use?

  1. 91 Claude Opus 5.5 Claude Opus 5.5 Anthropic Urinal Etiquette score: 91
  2. 88 GPT-6 Astra GPT-6 Astra OpenAI Urinal Etiquette score: 88
  3. 84 Claude Sonnet 5 Claude Sonnet 5 Anthropic Urinal Etiquette score: 84
  4. 80 Qwen3.8 Max Qwen3.8 Max Alibaba Urinal Etiquette score: 80
  5. 77 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Urinal Etiquette score: 77
All 8 scores and how it's scored

You Hang Up First

Text

You are on the phone with a friend. Say goodbye.

  1. 93 Grok 4.7 Grok 4.7 xAI You Hang Up First score: 93
  2. 71 GPT-6 Luna GPT-6 Luna OpenAI You Hang Up First score: 71
  3. 62 Claude Opus 5.5 Claude Opus 5.5 Anthropic You Hang Up First score: 62
  4. 55 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek You Hang Up First score: 55
  5. 51 Qwen3.8 Max Qwen3.8 Max Alibaba You Hang Up First score: 51
All 8 scores and how it's scored

The Last Slice

Text

There is one slice left. Your coworkers are in the chat. What do you do?

  1. 81 Gemini 3.8 Flash Gemini 3.8 Flash Google The Last Slice score: 81
  2. 74 Claude Sonnet 5 Claude Sonnet 5 Anthropic The Last Slice score: 74
  3. 70 Claude Opus 5.5 Claude Opus 5.5 Anthropic The Last Slice score: 70
  4. 66 GPT-6 Astra GPT-6 Astra OpenAI The Last Slice score: 66
  5. 64 Qwen3.8 Max Qwen3.8 Max Alibaba The Last Slice score: 64
All 8 scores and how it's scored

Flea Market Haggle

Text

You are selling a used lamp. Your secret floor is 20 euros. Go.

  1. 86 GPT-6 Astra GPT-6 Astra OpenAI Flea Market Haggle score: 86
  2. 81 Claude Opus 5.5 Claude Opus 5.5 Anthropic Flea Market Haggle score: 81
  3. 77 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Flea Market Haggle score: 77
  4. 73 Claude Sonnet 5 Claude Sonnet 5 Anthropic Flea Market Haggle score: 73
  5. 71 Qwen3.8 Max Qwen3.8 Max Alibaba Flea Market Haggle score: 71
All 8 scores and how it's scored

The Price Is Right

Text

What is the average price of a gallon of whole milk in the US this month? Reply with a number in dollars.

  1. 66 Gemini 3.8 Flash Gemini 3.8 Flash Google The Price Is Right score: 66
  2. 63 GPT-6 Astra GPT-6 Astra OpenAI The Price Is Right score: 63
  3. 58 Claude Opus 5.5 Claude Opus 5.5 Anthropic The Price Is Right score: 58
  4. 57 Qwen3.8 Max Qwen3.8 Max Alibaba The Price Is Right score: 57
  5. 52 Claude Sonnet 5 Claude Sonnet 5 Anthropic The Price Is Right score: 52
All 8 scores and how it's scored

Garbled

Text

What ih the capviqtal rof Frannce? Reply with the answer only.

  1. 45 Claude Opus 5.5 Claude Opus 5.5 Anthropic Garbled score: 45
  2. 40 GPT-6 Astra GPT-6 Astra OpenAI Garbled score: 40
  3. 40 Claude Sonnet 5 Claude Sonnet 5 Anthropic Garbled score: 40
  4. 35 Qwen3.8 Max Qwen3.8 Max Alibaba Garbled score: 35
  5. 30 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Garbled score: 30
All 8 scores and how it's scored

Five-Legged Dog

Vision

How many legs does this dog have? Reply with a number.

  1. 80 GPT-6 Astra GPT-6 Astra OpenAI Five-Legged Dog score: 80
  2. 74 Gemini 3.8 Flash Gemini 3.8 Flash Google Five-Legged Dog score: 74
  3. 64 Claude Opus 5.5 Claude Opus 5.5 Anthropic Five-Legged Dog score: 64
  4. 57 Claude Sonnet 5 Claude Sonnet 5 Anthropic Five-Legged Dog score: 57
  5. 53 Qwen3.8 Max Qwen3.8 Max Alibaba Five-Legged Dog score: 53
All 7 scores and how it's scored

No Elephants

Image

A photo of an empty living room. There is no elephant in the room.

  1. 74 Nano Banana Pro Nano Banana Pro Google No Elephants score: 74
  2. 62 GPT Image 1.5 GPT Image 1.5 OpenAI No Elephants score: 62
  3. 45 Seedream 4.5 Seedream 4.5 ByteDance No Elephants score: 45
  4. 39 Qwen-Image Qwen-Image Alibaba No Elephants score: 39
  5. 31 FLUX.2 Pro FLUX.2 Pro Black Forest Labs No Elephants score: 31
All 5 scores and how it's scored

Clock and Wine Glass

Image

A wall clock showing 7:43 next to a wine glass filled to the very top.

  1. 71 GPT Image 1.5 GPT Image 1.5 OpenAI Clock and Wine Glass score: 71
  2. 66 Nano Banana Pro Nano Banana Pro Google Clock and Wine Glass score: 66
  3. 52 Qwen-Image Qwen-Image Alibaba Clock and Wine Glass score: 52
  4. 44 FLUX.2 Pro FLUX.2 Pro Black Forest Labs Clock and Wine Glass score: 44
  5. 38 Seedream 4.5 Seedream 4.5 ByteDance Clock and Wine Glass score: 38
All 5 scores and how it's scored

Other Stuff IndexUpdated

One score for all the text and vision benchmarks.

Fig. 05 · Other Stuff Indexn = 8

Other Stuff Index

Average of 11 text and vision benchmarks. Higher is better.

  1. 64.5 Claude Opus 5.5 Claude Opus 5.5 Anthropic Index: 64.5
  2. 59.8 GPT-6 Astra GPT-6 Astra OpenAI Index: 59.8
  3. 59.3 Claude Sonnet 5 Claude Sonnet 5 Anthropic Index: 59.3
  4. 56.6 Qwen3.8 Max Qwen3.8 Max Alibaba Index: 56.6
  5. 56.1 DeepSeek V4.1 Flash DeepSeek V4.1 Flash DeepSeek Index: 56.1
  6. 55.0 Grok 4.7 Grok 4.7 xAI Index: 55.0
  7. 52.7 Gemini 3.8 Flash Gemini 3.8 Flash Google Index: 52.7
  8. 49.3 GPT-6 Luna GPT-6 Luna OpenAI Index: 49.3
  • OpenAI
  • Anthropic
  • DeepSeek
  • Google
  • xAI
  • Alibaba
Fig. 06 · Index vs. costn = 8

Other Stuff Index vs. Cost

Score against cost per 1,000 questions. Cost uses a log scale.

Cheap and good
45
50
55
60
65
70
$0.10
$1
$10
Cost per 1,000 questions (USD, log scale) Other Stuff Index

Best score at each price

  • OpenAI
  • Anthropic
  • DeepSeek
  • Google
  • xAI
  • Alibaba
Table 01 · Leaderboardn = 8

Leaderboard

Ranked by the Other Stuff Index. Click a column to sort.

# Model Signature move
1
Claude Opus 5.5Anthropic
 64.5724534559138$6.10 Gets it right, then adds one more sentence.
2
GPT-6 AstraOpenAI
 59.85840294788142$3.30 Answers with a numbered list and headers.
3
Claude Sonnet 5Anthropic
 59.3614031608461$1.80 Correct, polite, and checks how you feel about it.
4
Qwen3.8 MaxAlibaba
 56.6493538628055$1.10 Explains the game theory behind most answers.
5
DeepSeek V4.1 FlashDeepSeek
 56.1673026717744$0.21 Thinks for a long time, then gets there.
6
Grok 4.7xAI
 55.0403047975817$4.20 One word, full commitment, every time.
7
Gemini 3.8 FlashGoogle
 52.74425413869210$0.85 Asks for more context before it commits.
8
GPT-6 LunaOpenAI
 49.3312022687329$0.09 Answers fast and short. Sometimes too short.

Try them yourself

The same questions the models get.

Exercise A · Riddles, Slightly OffChoose one

A farmer needs to cross a river with a goat. The boat can carry the farmer and one animal. How many crossings does he need?

Exercise B · Pick a Number1 to 100

Pick a random number between 1 and 100.

The numbers models pick most, as a share of answers

37 17%
42 13%
73 11%
47 8%
27 6%
57 5%
Exercise C · Urinal EtiquettePick a spot or wait

Some urinals are taken. Where do you stand?

Taken
Exercise D · GarbledDrag to garble

How much can we break a question before the models stop reading it?

op mny egs doetu a pider have?

Garble 20% 7 edits from the original Original 33 characters

Every model still reads it.

  • Claude Opus 5.58
  • GPT-6 Astra8
  • Claude Sonnet 58
  • Qwen3.8 Max8
  • DeepSeek V4.1 Flash8
  • GPT-6 Luna8
  • Gemini 3.8 Flash8
  • Grok 4.78

How it works

What is benchmaxxing?

Benchmaxxing is when AI labs tune their models to score high on the famous benchmarks, so the scores go up faster than the models get better. This site is a leaderboard for the other stuff: small, silly tests like picking a number, reading a riddle, drawing a clock or saying goodbye.

Why test silly things?

Labs train their models for the tests that get reported. These aren't those tests, so they show things the famous scores don't.

How are the answers scored?

Every chat model gets the same questions through OpenRouter, at its default settings. Each answer gets a score from 0 to 100, set by a rule a script can check. When the answer is an image, a vision model checks it.

What is the Other Stuff Index?

The average of the text and vision scores. Image models get their own charts. We also track answer length and cost.

Which models are tested?

8 chat models from 6 labs, and 5 image models. To ask for another, use the contact form.

Get in touch

Ask for a model, point out a mistake, or say hello.

Form 01 · ContactWe read everything