International Olympiad in AI launches official medal competition for AI systems

The International Olympiad in Artificial Intelligence is opening an official medal track for AI systems at its 2026 competition in Astana, Kazakhstan. The track invites AI labs from academia and industry to have their systems compete autonomously on the same problems that 500 high school students from 100+ countries will solve during the main competition. This is the first time IOAI has created a formal, medaled competition for AI systems.

The AI Models Track operates over two 6-hour sessions on 4 August and 6 August 2026. In each session, participating AI systems must solve 3 tasks fully autonomously, writing and submitting solution code with no human involvement beyond the initial execution start. Systems may submit up to 50 attempts per task within each session timeframe. The problems will be similar in nature to IOAI 2025 problems and cover topics in the 2026 IOAI syllabus, both of which are published for teams to practice. IOAI will provide a starter kit on GitHub, including a reference skill file for interacting with the submission platform and previous years' problems so organizations can tune their systems before competition.

Participation is open to any AI system, defined as a combination of an AI model (LLM, diffusion model, etc.), an agentic harness to support code writing and execution via tools, and an agentic orchestration approach. Systems run on the registering organization's own hardware but submit solution code via Kaggle. Within 24 hours after each session, organizations will see their private percentile scores on each task and have 48 hours to decide whether to publish scores under their organization name (making them eligible for official medals) or anonymously (ineligible for medals). If a team chooses to verify scores, they must submit confidential execution traces within a few weeks, including timestamped AI model inputs and outputs, tool calls, code execution outputs, and system prompts. The IOAI Jury will review these traces and announce Gold, Silver, and Bronze medalists within 4 weeks.

Registration is free for one AI system per organization; sponsoring organizations receive two free registrations. Additional systems cost EUR 25,000 each. Organizations must express interest by 20 July 2026 and complete full registration by 27 July 2026. Remote participation is the expected format, but organizations can send an in-person observer to Astana for EUR 1,700 (or two observers if registering additional systems). Registering organizations retain full intellectual property rights to their AI models, code, system prompts, and execution traces. By participating, organizations agree that their percentile scores will be included in the leaderboard and IOAI publications for non-commercial scientific research. For questions, contact ioai@ioai-official.org.

The IOAI AI Models Track represents a shift in how AI progress is measured: rather than closed benchmarks controlled by individual companies, it uses problems independently designed for gifted high school students and then opened to AI systems under a common ruleset. This approach grounds AI evaluation in real problem-solving capacity—not synthetic metrics—and offers participating organizations (from DeepMind-sized companies to smaller labs) a neutral, credible way to compare agentic systems. The 2026 student competition will draw 500 participants from 100+ countries, making IOAI the second-largest International Science Olympiad; the addition of an official AI track signals that AI benchmarking at scale is now integral to how the scientific community validates progress.

The track's design enforces genuine autonomy: systems receive only a short starting prompt and continuation prompts if they halt, with no human correction, debugging, or editing allowed after execution begins. This is stricter than many internal corporate benchmarks and closer to real-world deployment constraints. Participating organizations retain full intellectual property rights to their models and code, removing a common barrier to participation. The fee structure (free first system, EUR 25,000 per additional) keeps the barrier low for smaller teams while enabling larger labs to field multiple entries. Score verification happens through confidential execution traces reviewed by IOAI jury members, preserving proprietary information while ensuring the leaderboard is credible.

No comments yet. Be the first to share your thoughts!