Vals AI Hackathon: Benchmaxxing
Vals AI is hosting its first hackathon in San Francisco, and for one weekend, benchmaxxing is the point.
Your challenge: post-train a small model to climb the leaderboards on three Vals benchmarks, or build a new eval that frontier models still can't solve.
We'll bring the compute, benchmarks, food, and prizes. You bring the ideas.
Two Tracks:
Post-Training: Post-train a small model and score it on three Vals benchmarks, using compute from our sponsors.
Build Your Own Eval: Build a benchmark on Valkyrie for a task frontier models still get wrong. We cover model and compute costs.
Who should apply?
Anyone can apply, but we give preference to applicants with technical experience.
Schedule
Saturday, November 14
12 PM: Check-in, snacks, and introductions
1 PM: Opening remarks
2 PM: Hacking begins
5 PM: Cybersecurity CTF (optional side quest)
7 PM: Dinner and guest speakers
10 PM: Dessert and show-and-tell
Sunday, November 15
9 AM: Breakfast
2 PM: Submissions close, demos and judging
3 PM: Winners announced
Prizes:
Prizes for 1st, 2nd, 3rd places for both tracks!
Speakers:
Do I need a team?
No, but we recommend one. Apply with a team of up to four, or apply solo and we'll match you with a team on-site.
Who is Vals AI?
Vals AI builds independent benchmarks that measure how well AI models perform on real professional work: law, finance, healthcare, coding, and frontier risks like cybersecurity and recursive self-improvement.
Public benchmarks saturate quickly, so we test capabilities the way they actually get used, and we keep our test sets private to keep the scores honest. We run every evaluation ourselves. This hackathon puts that work in your hands for a day.
For more info, visit vals.ai/hackathon
In-person, San Francisco. Approved applicants only, no door entry. Teams up to four. 18+.