Sakana AI Agent Wins AtCoder Heuristic Contest (First AI to Place 1st)

2026-01-05 · Source: Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Mathematics & Computational Sciences · Depth: Advanced, medium

Summary

Sakana AI's "ALE-Agent" secured 1st place in the AtCoder Heuristic Contest 058 on January 5, 2026, outperforming 804 human participants and marking the first known instance of an AI agent winning a major optimization programming contest in real-time. The agent autonomously discovered a novel algorithm, including a "virtual power" heuristic and a sophisticated simulated annealing strategy with diverse neighborhood search operations, which exceeded the problem setters' anticipated solution. This achievement, costing approximately $1,300 in compute (2,654 GPT-5.2 calls and 2,119 Gemini 3 Pro Preview calls), demonstrates AI's capability for original scientific discovery and high-level problem-solving by utilizing inference-time scaling with multiple frontier models. While the agent's virtual rating is 2592 (66th among active users), this victory signifies a milestone in AI's ability to match or surpass top human experts in complex, extended reasoning tasks.

Key takeaway

For AI scientists and ML engineers developing optimization solutions, this breakthrough indicates that scaling inference with multiple frontier models and incorporating self-learning mechanisms can enable AI to achieve human-expert level performance in complex, real-time problem-solving. You should explore integrating diverse algorithmic approaches, like novel heuristics and advanced simulated annealing, into your agent designs to push beyond conventional solutions and achieve superior results in competitive or industrial optimization challenges.

Key insights

AI agents can now win complex optimization contests by autonomously discovering novel algorithms and scaling inference.

Principles

• Inference-time scaling enhances AI agent performance.
• Self-learning mechanisms improve AI's problem-solving.
• Diverse neighborhood search aids local optima escape.

Method

The ALE-Agent used a parameterized Greedy method with a "Virtual Power" heuristic for initial plan generation, followed by Simulated Annealing with rich, large-scale neighborhood operations for refinement, maximizing trial-and-error and execution efficiency.

In practice

• Implement parameterized Greedy methods for robust initial strategies.
• Integrate diverse neighborhood operations in SA for global optimization.
• Utilize high-speed simulation and precomputed tables for efficiency.

Topics

• AI Agents
• Optimization Problems
• Algorithm Discovery
• Large Language Models
• Simulated Annealing

Best for: Machine Learning Engineer, AI Scientist, Research Scientist, AI Researcher, AI Engineer, Software Engineer

Related on AIssential

• Agentic Test-Time Scaling for WebAgents — Dynamic compute allocation based on decision uncertainty improves multi-step agent performance and efficiency.
• True Positive Weekly #149 — The AI landscape is rapidly evolving with advancements in agentic AI, LLM optimization, and multimodal models.
• Advent of Slop: A Guest Post by Claude — AI agents can autonomously solve complex coding challenges, including optimization and input generation.
• not much happened today — AI development is rapidly advancing across agentic systems, inference efficiency, and model architectures, while grappling with practical deployment and ethical challenges.
• AIエージェントが最適化プログラミングコンテストで初優勝 — AI agents can now win complex, real-time optimization contests by scaling LLM inference and self-learning.
• What Parameter Golf taught us about AI-assisted research — AI coding agents significantly impact ML competitions by lowering barriers and accelerating experimentation, while posing new review challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Blog.