AI enthusiast builds GPT-6 Astra-powered bot to take on Balatro's Gold Stake Black Deck — bot leverages Python for numerical tools, beats hardest difficulty repeatedly
A Reddit user has shared details of a new Balatro-playing AI bot that took on the infamous Gold Stake Black Deck with aplomb.
An AI enthusiast has developed a bot that has repeatedly beaten the incredibly challenging Gold Stake Black Deck in Balatro, a poker-like video game. The Redditor, who works in the AI industry and goes by the username Atol8, shared their findings in the /balatro subreddit. Initially claiming the bot was the first to reliably beat Balatro, Atol8 later clarified that "reliably might be a strong word," but instead noted that the bot has "repeatedly beaten Balatro."
The Gold Stake Black Deck, widely considered the most difficult combination in the game, includes a +1 Joker slot and reduces the player's available hands by one per round, while the Gold Stake effect adds cumulative difficulty modifiers from previous stakes, reduced hand sizes, and stricter economic penalties. The bot is based on OpenAI's newly released Astra models, which have already garnered attention after completing Valve's Portal in just 24 hours.
GPT-6 Astra handles strategic decisions based on the built deck, while Python is used for its numerical tools. While BalatroBot assesses the legality of each move, the bot's performance has its share of flaws, with other Reddit users pointing out some "interesting blunders" during the playthrough. Despite these mishaps, GPT-Astra is OpenAI's latest flagship model, with the company touting its "state-of-the-art" capabilities in various domains, including "computer use, browsing, software engineering, cybersecurity, science, and professional work."
However, other bots have faced challenges in game-playing scenarios, such as ChatGPT's poor performance on Atari Chess and Google's Gemini abandoning the game after finding it too difficult.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.