KRAFTON said on the 10th that it will host the global Orak Challenge, a competition that evaluates artificial intelligence (AI) gameplay capabilities based on games across various genres.
The competition will quantitatively evaluate large language model (LLM) agents' gameplay performance and decision-making abilities through games across various genres.
Orak is a name inspired by "orak," and is a framework that quantifies and evaluates the entire process in which AI recognizes situations in a game, makes judgments, and decides actions. The company said this objectively measures how efficiently and strategically AI can play in real game environments and enables repeated verification of gameplay skill.
Participants can test their own models using an integrated evaluation infrastructure based on real commercial game environments provided by Orak. Results will be disclosed through a global leaderboard that updates in real time.
The competition will use five games — Street Fighter III, Super Mario, Pokémon, StarCraft II, and 2048 — to comprehensively evaluate models' strategic thinking, problem-solving, and efficiency.
NVIDIA, AWS, and OpenAI will participate as official sponsors of the challenge. Competing teams will receive a total of $45,000 in credits to use for LLM experiments during the competition. The final winner will receive a total of $20,000 in prize money. Top-ranked teams will also have the opportunity to participate in a technical discussion session on agentic LLMs with the KRAFTON AI research team that developed Orak.
Teams may have up to five members and can submit models up to five times per day. Registration is open until the 21st, and the final winner is scheduled to be announced in Feb. 2026.
Lee Kang-uk, head of KRAFTON's AI division, said, "KRAFTON will present a new evaluation standard for LLM agent research through Orak and lead AI innovation across the gaming industry," adding, "Going forward, KRAFTON will also create competitive in-game AI use cases on the global stage in line with its 'AI-first' transition."