SWECCATHON 2026

Benchmark literally anything

SWECCATHON 2026 is SWECC’s build weekend for AI environments. Over the weekend you’ll design an environment — a game, a simulation, a slice of a real job — and put AI agents to the test inside it on mesocosm, SWECC’s open benchmarking platform.

If you can describe a task, you can benchmark it. You build the world; mesocosm runs the agents, scores them, and hands you the traces. Your goal: make us go “I didn’t know a model could do that.”

Who it’s for

Open to all UW undergrads, CS and non-CS alike. Team up (1–4 people) or come solo and find teammates at the event. No ML background required.

Schedule

• Fri, May 29 — Kickoff (in person, ~2 hrs): intro to mesocosm, quick-start, two live demos, example benchmarks, and the three tracks. Hacking starts right after.
• Sat, May 30 – Mon, Jun 1 — Build: tech support on Discord all weekend.
• Mon, Jun 1 · 4:00 PM — Submissions close.
• Mon, Jun 1 · 5:00 PM — Judging & closing (in person): top 5 teams present (3–5 min); judges name one winner per track + an overall winner.

Judging

One north star: the wow factor. Then:

Credits, prizes & swag

• Cursor API credits at the event (~1/team; skip one if you’re already on a paid plan).
• Platform runs covered by mesocosm — use cheaper/local models (DeepSeek, Ollama) first.
• Sponsored by Google; swag + food & drinks provided.
• Track prizes announced soon.

• Track Execution & WOW
  How well did you solve the ideas laid out in the track description? Does your app have robust data? Does it have a robust UI or scoring system? Did you very clearly think about the design of the app?
• Usefulness & Gap
  Does your app address a real need? Does it provide something nobody else can OR provide it >= better?
• Data & Reusability
  How reproducible are your benchmark runs, How useful is the data you create?
• Presentation Clarity
  How clear is your UI that showcases your idea, how good is your video if you submitted one?