Inaugural Season · Registration Opens TBD
AI MARATHON is the global build championship for the next generation of researchers, engineers, and founders. Form a team, pick a track, and turn an idea into a working AI system against the clock — judged by a panel of practitioners, evaluated on reproducibility, and celebrated on Demo Day.
Public criteria with published weights. Every finalist receives written feedback from the panel.
Submissions are evaluated on a sandboxed runner. If it doesn't run, it doesn't win.
Teams keep ownership of their work. We only ask for non-exclusive rights to showcase your demo.
Office hours with practitioners during the build window — not just a leaderboard, a community.
About AI MARATHON
AI MARATHON exists because the bar for what students and early-career builders can ship has changed. Foundation models, agent frameworks, and open weights put production-grade tooling in the hands of anyone willing to learn — but most competitions still optimize for puzzle-solving, not for building.
We're rebuilding the format from scratch. Real problems, real shipping, real evaluation. No private graders, no hidden rubrics. Submit code, get scored, get feedback. The teams that ship the most useful, most reproducible, most original work move on.
Lower the floor and raise the ceiling for the next generation of AI builders.
Open rules, open scoring, sandboxed evaluation, and a published code of conduct.
Working systems, public demos, mentor connections, and a portfolio piece that travels.
Competition Tracks
Every track shares the same evaluation rubric. Pick the one where your strongest idea lives.
Push novel research, reproducible benchmarks, or new training and evaluation methods. For teams who want to move the state of the art an inch forward.
Autonomous systems that complete real-world workflows end-to-end. Web, code, ops, research, scheduling — anywhere a human is pasting between tabs today.
Climate, accessibility, education, public health, civic infrastructure. Real users, real impact, measurable outcomes — judged on usefulness, not novelty alone.
The "anything that moves AI forward" track. Creative tools, developer experience, hardware, interfaces, weird ideas that wouldn't fit anywhere else. Surprise us.
How it works
Solo or up to four. Teammates can be anywhere in the world. We provide a teammate-finder for solo applicants who want to be matched.
Open build period with daily mentor office hours, async support, and clearly published rules on what's allowed (yes to foundation models, yes to open source, yes to LLM-assisted coding — with disclosure).
A reproducible repo and a 3-minute demo video. Submissions run on a sandboxed evaluator. Initial scoring is automated; final scoring is by the panel.
Top teams across all tracks present live. Public showcase, panel Q&A, and prizes announced. Every finalist receives a written rubric breakdown.
Timeline
Exact dates announced when registration opens. The build window is intentionally long enough to fit around school and work.
Apply with a sketch of your idea and your track. Teammate-finder opens for solo applicants.
TBDMulti-week open building. Daily mentor office hours. Rolling guidance posted in the public handbook.
TBDSandboxed evaluation + first-round panel review. Track-level finalists announced.
TBDFinalists present live. Public showcase. Awards announced. Written feedback delivered to every finalist team.
TBDRecognition
Final amounts announced at registration open. The structure is fixed and published in advance.
All cash amounts and credit totals are finalized and published before registration opens. AI MARATHON does not require participants to forfeit ownership of their submissions.
Judging Criteria
Is the idea genuinely new, or a meaningful re-combination? How far does it move beyond an obvious prompt or template?
Does the system actually work? Is the engineering sound? How robust is it under realistic inputs and adversarial cases?
Who benefits, and how much? Is the problem worth solving? Is there a credible path from prototype to people using it?
Can a stranger run your code and follow your demo? Is the README honest about limits? Is the writeup clear?
Eligibility, Rules & Code of Conduct
Organizing Committee
Final committee — including panel and advisory roles — is announced when registration opens. Slots below are placeholders until then.
Track lead · Frontier
Track lead · Agents
Track lead · AI for Good
Track lead · Open Innovation
Partnerships
We're inviting a small set of compute, community, and ecosystem partners for the inaugural season. Partner placements are auditable: we publish what every partner contributes, and we don't run pay-to-win categories.
Long-term naming relationship and Demo Day presence.
Cloud, GPU, or API credit pools for participants.
Universities, student orgs, and regional partners that host build hubs.
Partner inquiries get a same-week response with a one-page brief and tier sheet.
partners@aimarathonEmail is a placeholder until launch. Drop your interest via the registration form below and we'll follow up.
FAQ
Builders, researchers, and students who want a serious place to ship. We deliberately keep the floor low (no prerequisite courses, no entry fee, teammate-finder for solo applicants) and the ceiling high (reproducibility-graded, panel-judged, and Demo Day stage time).
No. Solo applicants can compete solo or opt in to the teammate-finder. Final teams of up to four can be re-formed during the registration window.
Yes — with disclosure. Submissions list every model, API, and tool used, including AI-assisted coding. Disclosure does not lower your score; non-disclosure is disqualifying.
No. AI MARATHON is free to apply and free to compete. If we ever introduce paid add-ons (e.g., extra compute), the free path will always remain competitive on its own.
Public criteria with published weights. Every entry passes an automated pre-screen, then receives at least two independent panel scores. Discrepancies above a threshold are escalated to a third panelist. Panel composition and conflicts are published.
You do. AI MARATHON asks only for a non-exclusive license to showcase your demo and writeup. We do not claim ownership of your code, models, or product.
Yes, if you're 13 or older. Minors need a guardian acknowledgment, which we'll guide you through during registration. Some prize disbursements may be routed through a guardian per applicable laws.
Three things: (1) form a team or sign up for the finder, (2) sketch one idea per track you'd consider, (3) skim the handbook the moment it drops. The handbook includes the full rubric, sandbox spec, and submission template.
Be first in line
Drop your details below and we'll send the handbook the moment it's public — plus first dibs on the teammate-finder and partner channels. No spam, unsubscribe anytime.
Your email is used only for AI MARATHON announcements. Full privacy policy published with the handbook.