GrandCode: First AI to Beat All Humans in Live Competitive Programming Contests

TLDR

GrandCode has become the first AI system to consistently beat all human participants in live competitive programming contests, placing first in three consecutive Codeforces competitions (Rounds 1087, 1088, and 1089 in March 2026). This surpasses previous benchmarks—Google's Gemini 3 Deep Think had only achieved 8th place.

The system employs multi-module orchestration coordinating hypothesis proposal, solver, test generator, and summarization components, all improved through post-training and online test-time reinforcement learning. A novel 'Agentic GRPO' training method addresses delayed rewards and off-policy drift common in agentic RL systems. The research demonstrates AI has reached a capability threshold for outperforming elite human programmers on highly competitive coding tasks.

Key Takeaways

• GrandCode is the first AI system to win live Codeforces competitions, placing first in three consecutive contests and outperforming legendary grandmasters using multi-agent reinforcement learning