Polish programmer 'Psyho' beats AI in 10‑hour marathon to win AtCoder World Tour Finals 2025 Heuristic Championship

Introduction

At the prestigious AtCoder World Tour Finals Heuristic Contest 2025 in Tokyo, Polish programmer Przemysław Dębiak (better known as “Psyho”) delivered a standout performance by defeating OpenAI’s custom AI model. Competing against the world’s top 12 human programmers and, for the first time, a direct AI contender, Dębiak demonstrated that human intelligence and creativity remain superior.

Table of Contents

Contest Format

Duration: 600 minutes (10 hours)
Challenge: Guide a robot across a 30×30 grid using the fewest possible steps
Participants: 12 elite human programmers and one OpenAI AI model named “OpenAIAHC”
Rules: All contestants used identical hardware without access to external libraries, internet, or documentation

The problem was NP-hard—lacking a clear “exact” solution—requiring contestants to find “good enough” solutions within a strict time limit.

Psyho’s Victory Story

Former OpenAI Officer

Dębiak, who previously worked at OpenAI, coded entirely as a human—using only Visual Studio Code and no AI helpers like Copilot .

Man vs. Machine

He said,

“I was so tired… but I was very close to getting a score comparable to the model… that pushed me to dig deep, using all the remaining energy.”

A Surging Finish

Although the AI led early on, Dębiak ultimately won with a margin of about 9.5%. The AI scored around 1.65 trillion, while Dębiak reached approximately 1.81 trillion .

Praise for the AI

Contest director Yoichi Iwata noted that while the AI excelled in optimization, it lagged in human creativity—and Dębiak outperformed it through inventive strategies .

Significance of the Win

Creative Advantage

While AI is fast, systematic, and strong at optimization, humans still have an edge in heuristic and creative problem-solving .

AI–Human Balance

Dębiak acknowledged:

“AI excels at fast, straightforward coding… but in longer contests… humans have edge.”

A John Henry Moment

The event has been likened to the legend of John Henry: human vs. machine, with human determination shining through in the era of mechanization.

Sam Altman & OpenAI’s Response

Sam Altman (CEO, OpenAI): Posted on X, “good job psyho,” acknowledging Dębiak’s victory .
OpenAI: Confirmed that their model took second place and congratulated Dębiak for keeping it at bay .

This event highlighted more than a single win—it spotlighted the enduring importance of human capability.

Looking Ahead

AI’s Rapid Progress: The Stanford AI Index 2025 shows AI-generated code success surged from 4.4% in 2023 to 71.7% in 2024 .
Realistic Humility: Dębiak remarked, “It’s easy to imagine a different problem where AI would win and all humans would be far behind.” .
Endurance Challenges: Long contests demanding strategy, stamina, creativity, and human insight remain tough for AI .

AI’s Rapid Progress

Benchmarks continue to showcase AI’s dramatic advancements. The Stanford AI Index 2025 reports sharp improvements—MMMU scores jumped by 18.8 points, GPQA by 48.9, and SWE‑bench by 67.3—indicating that in timed programming tasks, language-model agents can sometimes outperform humans .

Realistic Humility

Yet, human insight retains its unique value. Dębiak’s own reflections capture this tension: “It’s easy to imagine a different problem where AI would win and all humans would be far behind.” His point: while AI excels in brute-force approaches, humans bring nuance and adaptability .

Endurance Challenges

Sustained challenges—where strategy, fatigue management, creative improvisation, and broader context-awareness matter—pose distinct hurdles. AI often struggles in these extended, heuristic-based contests, underlining that long-haul mental resilience remains a domain where human programmers shine .

These themes suggest that the future lies not in humans competing against AI, but in hybrid frameworks where human creativity guides and complements machine speed—especially in complex, open-ended scenarios.

Conclusion

This contest transcended routine coding—it was a testament to human resolve, creativity, and physical-mental endurance. As AI continues to evolve, Dębiak’s victory is a potent reminder that in complex, creative, and time-pressured tasks, humans still lead.

However, this isn’t the end of the story—it signals the beginning of a new era of AI-human collaboration and competition, where every win, every loss, and every challenge carries profound meaning.