Why the arena closes
By the end of 2025, frontier models changed what a live leaderboard can prove. Results from the previous edition stand as the last public proof of skill for now.
Final leaderboardThe model problem
Late 2025 marked a real change in what frontier language models and coding agents could do. They became capable of solving non-trivial challenges and far more effective at brute-forcing their way through a problem until they found a working answer. More importantly for Alpha Challenge, these models have been trained on an expanding record of historical incidents and improved their tooling usage. One of our core principles is to use real onchain cases to teach applicable skills. When a model may already know the incident behind a challenge, it becomes much harder to distinguish genuine judgment and talent from data contamination and automated search.
What the arena proved
Alpha Challenge was built as an arena for talented people with high agency to prove what they could do. The leaderboard surfaced exceptional talent: some of its strongest performers went on to join research or DeFi teams at Wintermute, while others found opportunities at companies we connected them with.
What happens now
We do not want to run a 2026 competition whose leaderboard can no longer provide that same credible signal. Instead, a new batch of case studies is available for improving skills, together with previous editions. Work through them at your own pace. There is no registration, deadline, partner rewards, or new ranked competition this year; the 2025 leaderboard remains as the last record of the proof-of-skill.