An advanced artificial intelligence system developed by OpenAI has defeated a team of professional-level Dota 2 players in a high-profile exhibition, underscoring rapid progress in AI-driven decision-making.

According to reports, the AI—described as a GPT-5-powered evolution of earlier systems—secured convincing wins in two out of three matches against a lineup of experienced players, including Blitz, Cap, Merlini, MoonMeander and Fogged.

In the opening game, the AI established an early advantage through highly coordinated gameplay, forcing the human team to concede within 21 minutes. The second match proved more competitive initially, but the system again outperformed its opponents after the mid-game phase, sealing victory at around the 24-minute mark.

A third match, where spectators influenced the AI’s hero selection, saw the human side claim a win after the system was given a weaker lineup.

Players noted that the AI’s key strength lay in its near-instant decision-making and seamless team coordination—capabilities refined through extensive self-play training equivalent to decades of gameplay daily.

The demonstration builds on earlier milestones achieved by OpenAI’s Dota-playing systems, which have long been used to test artificial intelligence in complex, real-time environments. Such environments are considered critical benchmarks in the pursuit of more generalized AI systems capable of handling dynamic, multi-agent scenarios.

Analysts say the latest results highlight how AI systems are increasingly matching—and in some cases surpassing—human performance in strategic esports, raising broader questions about the future role of AI in competitive gaming and beyond.