THE CRUNCH

An AI enthusiast has built a bot powered by OpenAI's GPT-6 Astra that has repeatedly beaten Balatro's Gold Stake Black Deck, widely considered the game's hardest mode. The bot uses Python for numerical tools and BalatroBot to assess move legality. While the bot has made some blunders, it has successfully navigated the brutal economy and strict penalties of the Gold Stake.

The bot, created by Reddit user Atol8 (Jacopo Attolini), relies on GPT-6 Astra for strategic decisions and Python for numerical tools. A separate tool, BalatroBot, exposes the game state and controls for external programs. The bot's success comes despite the Gold Stake Black Deck being a combination of the most difficult modifiers in the game, including reduced hand sizes and stricter economic penalties.

Despite the bot's success, Reddit commenters have pointed out that it has made some 'interesting blunders' throughout its playthrough. This highlights that the bot is not playing a perfect game, even though it has repeatedly beaten the hardest difficulty. The bot's performance is a notable achievement for a large language model, which has historically struggled with game-playing tasks.

GPT-6 Astra is OpenAI's latest flagship model, which the company claims is state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work. The bot's success in Balatro suggests that the model's capabilities in strategic decision-making are improving. This is a significant step forward from previous models, which have struggled with game-playing tasks.

The bot's success is a testament to the power of combining large language models with traditional programming. While GPT-6 Astra handles the strategic decisions, Python and BalatroBot handle the numerical tools and move legality. This hybrid approach allows the bot to navigate the complex and difficult challenges of Balatro. The bot's success is a promising sign for the future of AI in gaming.

WHAT HAPPENS NEXT

It remains to be seen if other players will replicate this approach or if the bot will be improved to play more consistently.