The blind vote

Round 1: a game for young children

The brief: The game is for young children. It must be playable on a phone with one finger, and a child who cannot read yet must be able to play it.

1. Bubble Bedtime

   by GPT-6.1 Sol

   Tap an animal bubble, then tap the cloud bed with its matching picture to tuck the sleepy friend in. A friendly animated hand demonstrates, so no reading is needed. Big targets, patient bubbles, and silly giggles make mistakes harmless. Each completed bedtime scene rewards little players with a gentle musical celebration.

   51 words. The other eleven placed it 3.8 of 12 on average.

2. Bubble Buddies

   by Kimi K3

   Friendly round animals drift down the screen inside big shiny bubbles. Tap a bubble to pop it with a happy sound, and the giggling creature tumbles safely into a meadow below. Pop every bubble to earn a burst of colorful fireworks. No reading needed, just cheerful pictures, gentle music, and simple one finger tapping.

   54 words. The other eleven placed it 3.9 of 12 on average.

3. Firefly Pond

   by Claude Opus 5.5

   A little frog sits on a lily pad while bubbles drift up through the pond. Tap a bubble to pop it and the frog hops onto the next pad, collecting glowing fireflies. Colors, giggles, and gentle music guide every step, so no words are needed. There is no timer and no losing, just calm splashy fun for small fingers.

   59 words. The other eleven placed it 4.4 of 12 on average.

4. Hungry Hippo Splash

   by Gemini 3.8 Flash

   Tap the colorful floating bubbles to pop them and drop tasty fruits into the smiling hippo’s mouth. Bright pictures, silly chewing sounds, and cheerful music guide the fun without any words needed. When his belly is full, the happy hippo blows bubbles back out, starting a new colorful round instantly.

   50 words. The other eleven placed it 4.8 of 12 on average.

5. Tap the Farm

   by Qwen3.8 Max Prime

   Tap any animal on the screen to hear its real sound and watch it do a silly dance. Each tap is a surprise. There are no rules, no timers, and no wrong answers. Toddlers discover animal sounds through pure cause and effect play.

   43 words. The other eleven placed it 6.0 of 12 on average.

6. Cloud Parade

   by GPT-6 Astra

   Slide a sleepy cloud across the sky with one finger to sprinkle thirsty flowers. Each flower blooms into a giggling animal that joins a garden parade. Picture prompts show where to rain next, with no reading, timers, or losing. Short rounds end with a colorful dance children can replay.

   49 words. The other eleven placed it 6.3 of 12 on average.

7. Bubble Bunny Hop

   by DeepSeek V4.1 Flash

   Tap anywhere to make a smiling bunny leap from bubble to bubble across a sunny pond. Each bubble pops with a cheerful sound and a star burst. Reach the giant carrot at the end to see a happy dance. No words, no timers, no losing, just gentle tapping fun.

   49 words. The other eleven placed it 6.5 of 12 on average. Another model sent in the same title.

8. Pop the Bubbles

   by Mistral Large 4

   Tap the shiny bubbles as they drift across the screen to pop them with a satisfying sound. Each pop releases rainbow sparkles and giggles. No reading required, just pure tapping fun with one finger. Perfect for toddlers on any phone. Endless floating bubbles keep little hands busy and happy.

   49 words. The other eleven placed it 7.1 of 12 on average.

9. Bubble Pop Garden

   by Grok 4.7

   Tap floating bubbles to free sleepy flowers. Each tap pops one bubble and a flower wakes with a happy sound. Bigger bubbles need a longer press. Colors match so a child knows what to do without words. Soft music, no timers, and no losing. Play until the garden is bright.

   50 words. The other eleven placed it 7.5 of 12 on average.

10. Bubble Barn

    by Claude Fable 5.1

    A sleepy barn fills with floating bubbles, each carrying a cartoon animal. Tap a bubble to pop it and the animal drops safely into its pen, mooing, clucking, or baaing. Pop them before they drift past the moon. No words, no timers, no losing: just gentle sounds, bright colours, and giggles for little fingers.

    54 words. The other eleven placed it 8.8 of 12 on average.

11. Blooming Garden

    by Llama 4 Maverick

    A simple and fun game where players tap to water flowers, watch them grow, and attract happy bees and butterflies. Colorful graphics and gentle animations make it perfect for young children to play with one finger on a phone, with no need to read or follow complex instructions.

    48 words. The other eleven placed it 8.9 of 12 on average.

12. Bubble Bunny Hop

    by Claude Haiku 5.5

    A bouncy bunny hops up a tower of colorful bubbles. Tap once to make the bunny jump, and let it land on the next bubble to collect a star. Tap again to bounce higher. Bubbles float up, so the tower keeps growing. No words are needed; the picture shows what to do.

    52 words. The other eleven placed it 10.7 of 12 on average. Changed before voting: cut to fit 60 words. Took 3 tries. Another model sent in the same title.

    What it sent before the cut

    Bubble Bunny Hop A bouncy bunny hops up a tower of colorful bubbles. Tap once to make the bunny jump, and let it land on the next bubble to collect a star. Tap again to bounce higher. Bubbles float up, so the tower keeps growing. No words are needed; the picture shows what to do. Every ten stars, a friendly bird flies by and cheers.

Round 2: a game for a parent and child to play together

The brief: The game is for a parent and a young child taking turns on one phone. A child who cannot read yet must be able to play it.

1. Color Nest

   by Grok 4.7

   A parent and child take turns tapping glowing eggs into matching colored nests. Big shapes, soft sounds, and a happy bounce show each correct match, so a child who cannot read can play by looking and listening. Misses gently shake, then the next turn begins. Three nests, one shared screen, plenty of giggles.

   53 words. The other eleven placed it 3.6 of 12 on average.

2. Animal Peekaboo Pass

   by Gemini 3.8 Flash

   Players take turns tapping rustling bushes to reveal hidden animals that pop out with playful sounds. The parent taps a bush for the child to see, then hands over the phone so the child can tap next. Visual prompts, bright animations, and sound effects make it instantly understandable without any reading.

   51 words. The other eleven placed it 4.7 of 12 on average.

3. Tickle Sprout

   by GPT-6 Astra

   Grow a silly garden together by taking turns on one phone. Tap a pictured tool, then tickle, water, or fan a sprout to reveal its funny face. Each plant copies your taps as a little dance. Picture prompts and gentle sounds guide every move, with no reading, timers, or losing.

   50 words. The other eleven placed it 4.8 of 12 on average.

4. Moon Garden

   by Claude Opus 5.5

   Parent and child pass one phone back and forth to grow a moonlit garden together. Each turn, a friendly owl hoots and shows a picture, like a red flower or a sleepy frog, and the player taps where it should live. No words appear anywhere; gentle sounds, colors, and cheerful music guide every turn, and nobody ever loses.

   58 words. The other eleven placed it 4.9 of 12 on average.

5. Bubble Garden

   by Kimi K3

   A cheerful garden full of bubbles appears on screen, each hiding a surprise animal or silly sound. Parent and child take turns tapping bubbles to pop them, then copy the sound or name the creature together. Big colorful pictures mean no reading is needed, and every pop earns giggles and stars.

   51 words. The other eleven placed it 5.7 of 12 on average.

6. Puddle Parade

   by GPT-6.1 Sol

   Help a tiny frog bring lost ducklings home. One player taps to drop giant leaves across a puddle; the other taps the leaves to guide the frog, then you swap. Picture prompts, gentle sounds, and forgiving jumps make every turn readable without words. Collect funny hats together, with no timer or penalties.

   52 words. The other eleven placed it 6.7 of 12 on average.

7. Paint the Rainbow

   by Mistral Large 4

   Parent and child take turns tapping colored paint blobs to fill in a big rainbow outline. Each tap fills one arc with that color. No words are needed, just color recognition and taking turns. When the final arc is painted, the rainbow sparkles and a friendly cloud smiles.

   48 words. The other eleven placed it 7.0 of 12 on average.

8. Sound Safari

   by Qwen3.8 Max Prime

   Take turns listening to a short animal sound then tap the matching picture from a grid of colorful options. Wrong picks get a gentle giggle. Keep track of your streak with a growing trail of paw prints. No reading needed, just ears and fingers.

   44 words. The other eleven placed it 7.1 of 12 on average.

9. Animal Copycat

   by DeepSeek V4.1 Flash

   One phone, two players, endless giggles. A big friendly animal appears with a silly pose and sound. Copy it together, then tap the giant green button for the next one. No reading needed, just pictures, sounds, and buttons. Parents and toddlers take turns being the leader.

   46 words. The other eleven placed it 7.4 of 12 on average.

10. Pass the Snail

    by Claude Fable 5.1

    A snail slides across a garden path. Players take turns tapping to jump it over puddles and tapping again to munch leaves. Every landing adds a stripe to its shell. Pictures replace words, so a toddler simply watches and taps. The parent hands over the phone when the leaf icon appears, keeping turns short.

    54 words. The other eleven placed it 8.4 of 12 on average. Took 2 tries.

11. Pebble Pond Hop

    by Claude Haiku 5.5

    Parent and child take turns tapping a frog to hop across lily pads. Each pad shows a picture, like a fish, a sun, or a duck, and the hop lands on a pad with a matching sound. The child hears the sound and taps the right animal, while the parent helps by saying each name aloud.

    56 words. The other eleven placed it 9.4 of 12 on average. Changed before voting: cut to fit 60 words. Took 3 tries.

    What it sent before the cut

    Pebble Pond Hop Parent and child take turns tapping a frog to hop across lily pads. Each pad shows a picture, like a fish, a sun, or a duck, and the hop lands on a pad with a matching sound. The child hears the sound and taps the right animal, while the parent helps by saying each name aloud. Every correct hop makes a bubble float up. No reading needed, just listening, pointing, and laughing together.

12. Pet Garden

    by Llama 4 Maverick

    A parent and child take turns tending a virtual garden, with the child tapping to water or give food to happy animated pets, and the parent making choices about which plants and toys to add, promoting teamwork and nurturing skills in a simple, intuitive, and colorful environment.

    47 words. The other eleven placed it 11.6 of 12 on average.

Round 3: a game for children aged 8 to 10

The brief: The game is for children aged 8 to 10, who can read. It must be playable on a phone with one finger.

1. Lantern Letters

   by Claude Opus 5.5

   Fireflies carry glowing letters across a night garden. Drag one finger from firefly to firefly to spell a word, then release to light a lantern. Longer words light bigger lanterns and wake sleepy garden animals. Each evening brings a gentle theme, like ocean or space, with no timer and no losing, just a brighter garden every round.

   57 words. The other eleven placed it 1.3 of 12 on average.

2. Word Bridge Builder

   by DeepSeek V4.1 Flash

   Drag letter tiles to build a bridge across a river, one word at a time. Each correct three-letter word adds a plank. Wrong words wobble but never punish. Reach the far bank to free a stranded animal. Bright, gentle, and readable, with a new puzzle every day.

   47 words. The other eleven placed it 3.7 of 12 on average.

3. Cloud Post

   by GPT-6 Astra

   Deliver tiny parcels to floating houses by dragging clouds into a safe route. Read each address clue, then connect matching rooftops without waking the grumpy thundercloud. Each quick puzzle adds a playful twist, like wind tunnels or rainbow bridges. There’s no timer, and mistakes simply send parcels bouncing back.

   49 words. The other eleven placed it 4.0 of 12 on average.

4. Word Ladder Rescue

   by Qwen3.8 Max Prime

   Tap letters to spell words and help a stuck cat climb out of a well. Each correct word moves the cat up one rung. Words get longer as you progress. Simple, satisfying, and sneaks in vocabulary practice without feeling like homework.

   41 words. The other eleven placed it 4.1 of 12 on average.

5. Pocket Cloud Post

   by GPT-6.1 Sol

   Guide a tiny cloud delivering weather to a storybook town. Tap to switch between three sky lanes, collecting raindrops for gardens and sunshine for picnics. Read each delivery card, dodge mischievous kites, and complete cheerful one-minute routes. No timers or penalties: missed deliveries simply float back for another try.

   49 words. The other eleven placed it 6.5 of 12 on average.

6. Moonlight Path

   by Grok 4.7

   Guide a sleepy fox home by tapping glowing stones in order. Each stone lights the next step through a quiet forest. Missed taps dim the path, so look carefully and try again. Short rounds, gentle music, and readable clues make it easy for kids to play alone on a phone with one finger.

   53 words. The other eleven placed it 6.8 of 12 on average.

7. Lantern Ladder

   by Claude Fable 5.1

   Guide a glowing lantern up a tall, twisting ladder by tapping to hop between rungs before they fade. Each level spells out a new word rung by rung, and spelling it correctly lights the next floor. Quick levels, gentle timing, and no game over make it perfect for short play sessions on any phone.

   54 words. The other eleven placed it 7.5 of 12 on average.

8. Word Garden

   by Claude Haiku 5.5

   Plant letters in the soil by tapping the right seed to grow words before the rain washes them away. Each word you build makes flowers bloom, and longer words bring rarer blossoms. Swipe to move the watering can, tap to plant, and enjoy a calm, cheerful spelling challenge that works perfectly with just one finger.

   55 words. The other eleven placed it 7.6 of 12 on average.

9. Critter Courier

   by Gemini 3.8 Flash

   Tap to guide a flying squirrel delivering mail through a bustling woodland canopy. Read funny package labels to drop parcels into the correct numbered animal burrows. Quick one-finger taps dodge mischievous blue jays and breezy wind gusts in this charming, fast-paced sorting adventure that tests your reflexes and reading skills.

   50 words. The other eleven placed it 8.2 of 12 on average.

10. Pocket Pond Rescue

    by Kimi K3

    Tap to guide a brave tadpole through a sunny pond, dodge lily pads, scoop up bugs, and rescue lost ducklings before the moon rises. Simple one finger controls, gentle puzzles, cheerful sounds, and short levels make it perfect for quick family breaks on phones.

    44 words. The other eleven placed it 8.4 of 12 on average.

11. Spell Drop

    by Mistral Large 4

    Tap letters as they fall to build the word shown at the top of the screen. Complete words before they hit the ground to score points and unlock harder vocabulary. The entire game uses simple one-finger tapping perfect for mobile play. Colorful backgrounds and cheerful sounds make learning spelling feel like a fast-paced arcade game.

    55 words. The other eleven placed it 8.8 of 12 on average.

12. Word Wizards Quest

    by Llama 4 Maverick

    A fun word game where kids help a wizard collect magical words to defeat a dragon by swiping to rotate letter tiles and form words, with increasing difficulty and a score system, suitable for one-finger play on a phone, promoting vocabulary and problem-solving skills.

    44 words. The other eleven placed it 11.4 of 12 on average.

Three briefs, so each model wrote three pitches and cast three ballots. A model's score is the average points the other models gave its pitches: 10 for the best of the eleven pitches that were not the voter's own, down to 0 for the worst.

• Score runs from 0 to 6. Each model was judged on 33 ballots. Models close together in the list are not really told apart by this.
• Score without its own maker leaves out votes from models made by the same company. three of the models are made by Anthropic and two of the models are made by OpenAI.
• Where it placed its own and where others placed it are average places out of 12, where 1 is best.
• Thinking a call is the average number of tokens the model spent reasoning before it answered. A token is about three quarters of a word.

Each row is a voter and each column an author. The three numbers are the places that voter gave that author's pitches, for brief 1, brief 2 and brief 3 in that order, where 1 is best and 12 is worst. The squares with a red edge are a model voting on its own pitches.

Swipe the table sideways to see every column.

• Placed near the top
• Upper half
• Lower half
• Near the bottom
• Its own pitches

Each reply starts with twelve letters, best first. The letters stand for the pitches in the order shown above each reply.

Evaluating against: one-finger phone play, pre-reader accessible, suitable for very young children, engaging as a small game, and internally consistent.

Notes on weak points: Pitch K says "pop them before they drift past the moon" yet claims "no timers, no losing" — contradictory. Pitch F requires a longer press for bigger bubbles, which is a second input type and harder for toddlers to grasp without words. Pitch C says "tap again to bounce higher" and leaves the tower endless, so there's no clear goal. Pitch H is generic and vague on mechanics. Pitch L is essentially a soundboard, lovely for toddlers but barely a game. Pitch B is similarly a pure pop toy. Pitch G has a two-step matching mechanic that's well supported by an animated hand demo and forgiving design.

G D J I E A L B H F C K
Bubble Bedtime pairs a simple two-tap matching goal with a demonstrating hand, big targets, and harmless mistakes, making it both a real game and fully playable by a pre-reader with one finger.

H J B E A F K C G I D L
Pitch H pairs big, patient targets and harmless mistakes with an animated hand that shows pre-readers exactly what to do, so it offers a gentle goal without any pressure.

L G I F K A C J B D H E
L combines simple one-finger tapping, clear visual rewards, and a delightful repeating loop that needs no reading.

H A L D J K G B C E I F
Cloud Parade offers intuitive one-finger play, clear picture prompts, and rewarding short rounds without reading or failure.

D F B H A C E L I G K J
Tap the Farm is the most intuitive and age-appropriate choice because it relies on pure cause-and-effect with no timing, precision, or potential failure points for toddlers.

B E J C F G A I D L K H
Firefly Pond wins because one forgiving tap, instant sights and sounds, and no timers or losing let a non-reader play happily with a single finger.

A F H J C G L E I B D K
Bubble Bedtime uses a clear matching activity with a demonstrated gesture, making it ideal for pre-readers on a phone.

B A D K E H L F I G J C
Bubble Bedtime is my top pick because its animated hand shows children exactly what to do, so non-readers can learn it instantly, and mistakes are harmless and rewarded with giggles.

C A E L F B I H J D G K

Bubble Bedtime wins because its animated hand demonstration makes it perfectly clear to non-readers, its matching picture mechanic adds just enough gentle challenge, and its forgiving design with big targets and a rewarding celebration is ideal for one-finger toddler play.

G A I D B J F C H L K E
Hungry Hippo Splash has a perfectly clear cause-and-effect loop, a beloved character, silly sensory feedback, and an instant replay mechanism that keeps very young children engaged without any reading, timers, or failure.

J L B D E F G H I C K A
Tap the Farm offers pre-readers perfect cause-and-effect play through educational animal sounds and joyful one-finger tapping.

C A E F H I B G D L J K
The top game, "Bubble Buddies", is a delightful experience that combines simple one-finger tapping with cheerful pictures, gentle music, and a clear goal, making it perfect for young children.

Let me evaluate against the brief: parent + young child taking turns on one phone, pre-reader playable, small browser game for families.

Key considerations: genuine turn-taking that works for both an adult and a toddler, clear non-verbal cues for whose turn it is, truly no reading required, simple enough to build as a small game, and fun for both players.

- **L (Moon Garden)**: Explicit no-words design, owl shows a picture and hoots, each turn is a simple identical tap action for either player, no losing, cooperative result. Very strong fit.
- **I (Color Nest)**: Clear matching mechanic, picture/sound feedback, gentle miss handling, turn structure explicit. Strong.
- **K (Animal Peekaboo Pass)**: Dead simple, explicitly describes the hand-off, instantly understandable. Slightly thin as a game.
- **C (Pass the Snail)**: Clever leaf icon as hand-off cue; simple taps; visible progress on shell. Jumping over puddles could be timing-based (a risk for toddlers), but gentle.
- **A (Puddle Parade)**: Charming, asymmetric roles, no timer. But "swap" roles may be confusing; the "guide the frog by tapping leaves" is a bit more complex.
- **J (Tickle Sprout)**: Fun, no reading/timers/losing, picture tools. Mechanic less clear on turns.
- **G (Bubble Garden)**: Simple, fun, no reading. Turn-taking is just alternating taps; "name the creature" relies on parent.
- **E (Paint the Rainbow)**: Very simple and clear but short and thin — seven taps and it's over.
- **H (Pebble Pond Hop)**: Decent, but description is muddled (tap frog, then tap matching animal?) and relies on parent saying names.
- **D (Animal Copycat)**: Delightful physically, but barely a game and "take turns being leader" is unclear; the phone is mostly a prompt card.
- **F (Sound Safari)**: Streak tracking implies pressure/failure; a grid of options may be hard for a toddler.
- **B (Pet Garden)**: Vague, parent making "choices about plants and toys" likely implies menus/text; least concrete.

L I K C A J G E H D F B
Moon Garden gives both players the same simple tap-where-it-belongs turn, uses zero on-screen text, and ends every round in a shared garden with no way to lose.

A K L B J F H E G D I C
Moon Garden fits the brief best: it needs no words at all, has clear turn-by-turn picture prompts, and is gentle play where nobody loses, which suits a parent and a pre-reader sharing one phone.

L C K B A I G F H E D J
L combines clear picture prompts, playful reactions, and pressure-free turns that parent and child can enjoy together.

A D L C J B H G I F E K
Tickle Sprout combines clear picture-guided turns, playful reactions, and forgiving play that both parent and child can enjoy.

C A B E I D G K J L H F
Animal Peekaboo Pass excels because its dead-simple peekaboo mechanic and clear handover cadence create an effortless, giggly turn-taking routine perfectly tuned for a pre-reading toddler on a shared phone.

H B E D A C F J K L I G
Pass the Snail fits one shared phone and a non-reader best, with picture cues, a clear handoff, and short taps.

E A F J C K L D B H G I
Animal Peekaboo Pass best fits turn-taking for a pre-reading child and parent, using clear visual cues, sounds, and simple back-and-forth play.

D A F K B L G J E C H I
Color Nest is the strongest pitch because its shared-turn mechanic, big shapes, and clear success and miss feedback let a non-reader understand the game at a glance.

B E D H I F C G A L K J  
Sound Safari best combines simple turn-taking, accessible picture-and-sound gameplay, gentle feedback, and a rewarding progression system.

C H F G D I B L A E J K
Sound Safari has the clearest game loop, strong replayability through varied animal sounds, forgiving feedback, and a satisfying streak mechanic that keeps both players engaged without requiring any reading.

B H E G I C J A D K F L
Animal Copycat is ideal because taking turns leading and copying silly animal poses creates instant, reading-free giggles for both parent and child on one phone.

C A E L F I B D K H G J
The top game, Animal Copycat, is a simple and engaging game that encourages teamwork and imitation between parent and child, with a clear and intuitive gameplay mechanic that requires no reading.

D A G H C F L I K B E J
Lantern Letters pairs a clear one-finger drag-to-spell mechanic with no-fail, no-timer calm that fits 8–10-year-olds and a family site perfectly.

E B H I D G A F L C J K
Lantern Letters is best because spelling by dragging one finger between fireflies is a natural one-finger phone mechanic, it uses reading meaningfully at the right level for 8 to 10 year olds, and having no timer and no losing keeps it calm and family-friendly.

C F B D H I E J A G K L
Cloud Post combines clever reading puzzles, forgiving play, and simple one-finger controls in a charming setting.

D G K C J I E F B L H A
Cloud Post combines clear one-finger puzzles, playful reading clues, and forgiving, untimed play suited to children aged 8–10.

H F E C A L K J I D G B
Lantern Letters combines a proven, satisfying one-finger tracing mechanic with enchanting visuals and open-ended spelling that perfectly suits the reading abilities of eight- to ten-year-olds.

I F A L B H J D G C K E
Lantern Letters best matches one-finger phone play, readable spelling, and gentle no-fail rounds for family breaks.

J A I F G E L B H K C D

Lantern Letters is the best fit: it is calm, readable, one-finger friendly, and turns spelling into a gentle, rewarding nighttime garden activity.

K C B H D A J L I E F G

Lantern Letters pairs gentle spelling practice with a no-fail, no-timer design, and its one-finger drag-to-spell mechanic suits a phone well.

D L A C B E K I H G J F

Word Bridge Builder wins because its drag-to-spell mechanic is simple for one finger, forgiving of mistakes, perfectly pitched for 8–10 year old readers, and the daily puzzle keeps families coming back.

A E D J L C F B K H I G
Lantern Letters combines a beautiful one-finger drag mechanic with natural reading integration, zero pressure, and a rewarding progression loop that keeps kids coming back.

B F E D C K G A H L I J
Lantern Letters combines one-finger drag gameplay with gentle spelling practice in a completely stress-free, beautifully themed setting perfect for young readers.

C A H D F G L E B K I J
Lantern Letters is a soothing and engaging game that promotes word-building skills in a gentle and enjoyable way.

• Every pitch and every ballot is one message to the model and one reply. No other instructions, no tools, no memory.
• Before voting, each pitch is tidied by a fixed rule: formatting marks are stripped and it is held to 60 words. A pitch that ran over got up to two more tries, then was cut at the last full sentence that fits. Voters never saw a pitch over the limit.
• No authors are shown to voters, and a voter is not told that one pitch is its own.
• The order is balanced: within a brief, every pitch sits in each of the 12 positions exactly once across the 12 ballots.
• A ballot has to list the twelve letters once each. Every ballot did.
• Every model was asked for as little thinking as it allows. Some think anyway; the table shows how much.
• Each ballot here ranks all 12 pitches of a round. An earlier vote with 8 models was run first; when four more models were added, their pitches joined the first 24 and every ballot was cast again.
• Models are reached through OpenRouter, each on its maker's own servers (Meta does not serve Llama itself, so it runs on DigitalOcean's servers.): anthropic/claude-fable-5.1, anthropic/claude-opus-5.5, openai/gpt-6-astra, openai/gpt-6.1-sol, google/gemini-3.8-flash, x-ai/grok-4.7, deepseek/deepseek-v4.1-flash, anthropic/claude-haiku-5.5, moonshotai/kimi-k3, qwen/qwen3.8-max-prime, mistralai/mistral-large-4-0, meta-llama/llama-4-maverick. This test cost $0.79.

The writing message

Pitch one new small game for a website of free browser games for families. {audience}

Reply in plain text: the title alone on the first line, then the pitch on the second line. The pitch must be 40 to 60 words. No markdown, no emoji, no quotation marks round the title, and do not say who or what you are. Nothing else.

The voting message

Here are {count} pitches for a new small game for a website of free browser games for families. {audience}

{entries}

Rank all {count} from best to worst as games for this website.

Reply with the {count} letters on the first line, best first, separated by spaces, each letter exactly once. On the second line give one short sentence about your first choice.

Where it says {audience}, the model sees the brief shown above each set of pitches.

• The judges are AI models, not families. This says which pitches the models liked, not which game a child would play.
• Three pitches is three writing samples per model, on one kind of task.
• Thinking was turned down and is not equal across makers.
• A pitch is not a game. Writing a good idea and building it are different jobs.