Sloth in the City

Start your journey in central park beneath the canopy
Climb trees and explore their branches to forage
Cruise through the park in a park services cart
Row a boat across the lake and make new friends that join you on your journey
Take the subway on realistic routes with transfers
Stand clear of the closing doors when it's not your stop
Get to the Bronx and make your way to the zoo
Hop on a skateboard to zip around the zoo
Complete zoo quests to add to your menagerie and re-unite with your sloth friends
Pick the lock using your claws
Bring the whole crew back to the shuttle to the museum
Cruise down a realistic route dodging traffic along the way
Explore the museum with your friends on electric scooters
Show them the ancient ground sloth Megatherium for the finale

Inspiration

Inspired by 3d world exploration games and the groundbreaking capabilities of OpenAI's 5.6 Sol model for development and 3d modeling, as well as a deep love for learning, New York City and sloths.

What it does

Traverse NYC beginning in Central Park and traveling through Manhattan up to the Bronx Zoo, learning facts from your field guide about the city and the animals around you along the way, you re-unite with your sloth companions (and optionally a growing menagerie of other animals through various challenges) to shuttle them back to the the American Museum of Natural History to learn about the ancient ground sloth Megatherium. Throughout you'll take various modes of transit including the subway, a park services golf cart, a skateboard, shuttle bus (driving), and scooters and meet all sorts of interesting furry and feathered friends.

How we built it

Built entirely via ChatGPT Codex using GPT 5.6 Sol, leveraging its computer use capability to operate Blender to generate 3d character models and coordinate sound effect creation, OpenAI's Image 2 generation for textures and in-game imagery.

Challenges we ran into

As the game grew, QA became increasingly painstaking, but by adding teleportable checkpoints and leveraging computer use to have Codex play the game the primary limits have just been input ideas and time to cook.

Accomplishments that we're proud of

World modeling (especially Central Park, Bronx Zoo, and the highway level), cohesive computer use coordinated sound effects and engaging gameplay, and overall end to end output.

What we learned

Trying to build and QA too much in a single Codex session is unideal, but by splitting into separate sessions and parallelizing workstreams the sky is the limit.