Fun
Benchmark
Discover

The Laboratory

The goal: Brain Candy Studios is a small software and AI engineering shop and benchmarking lab. It exists to create fun, challenging, satisfying, and replayable games, and to use those games as the basis for benchmarking AI models. We share the benchmark data and what we learn with the community so others can inspect, reproduce, and build on the results.

Three pillars

Challenging fun. Build games that are enjoyable, demanding, satisfying, and worth playing again.

Capability-test the models. Use real games and real engineering work to test what AI models can do, where they struggle, and how they differ.

Discover the patterns. Track results over time to find the patterns behind changing capabilities. Publish those results along with our methodologies and tools.

The questions

We start by asking questions like:

• How can you define fun?
• Can an AI model predict whether humans will find something fun or challenging?
• Can an AI model independently create something fun and challenging? If not, what is stopping it?
• Can AI models play the game better than humans? Where are the gaps?
• Why does Model A perform Task 1 better than Model B, while Model A cannot do Task 2 at all and Model B struggles through it?
• How do these capabilities change over time?

As we seek answers to these questions and more, we aim to discover what patterns emerge over time.