Great progress for xAI: Grok is now delivering strong answers on Baldur’s Gate, a key focus area for Elon Musk’s AI venture. While different AI labs prioritize distinct goals—OpenAI with consumer applications, Anthropic with enterprise tools—xAI has made clear its interest in mastering niche, complex domains, particularly video game walkthroughs. This focus came into sharp relief in a recent Business Insider report by Grace Kay, which detailed how Musk personally intervened in the development of Grok, delaying a model release for days over dissatisfaction with its responses about Baldur’s Gate. According to sources familiar with the matter, high-level engineers were pulled from other projects to refine Grok’s ability to answer detailed questions about the game. The anecdote underscores Musk’s hands-on approach and the lengths he’s willing to go to ensure Grok excels in specific, often unexpected areas. While such demands might frustrate seasoned engineers used to tackling abstract problems in machine intelligence, they also reveal a clear strategic intent: to build a model that performs exceptionally well in real-world, practical contexts. To test whether this effort paid off, our resident RPG enthusiast Ram Iyer designed a five-question benchmark—dubbed BaldurBench—focused on core aspects of Baldur’s Gate gameplay. The responses from Grok were compared to those of the leading models: ChatGPT, Claude, and Gemini. The results were encouraging. Grok provided detailed, accurate, and technically sound answers, though its use of gaming jargon—such as “save-scumming” instead of “saving” and “DPS” rather than “damage”—could be confusing for newcomers. It also leaned heavily on tables and intricate theorycrafting, consistent with the expectations of hardcore players. Stylistically, the models varied: ChatGPT favored concise bullet points and fragmented sentences, while Gemini emphasized bolded keywords for emphasis. Claude stood out for its caution—when asked about optimal party builds, it concluded with a gentle reminder to “not stress too much and just play what sounds fun to you,” reflecting its broader design philosophy of preserving user experience. Importantly, this test area aligns directly with xAI’s known focus. The reported engineering sprint to improve Grok’s game knowledge wasn’t just a quirky anecdote—it was a deliberate effort to match or exceed competitors in a specific domain. Given that, it’s not surprising that Grok’s performance now holds its own against the others. While the results don’t suggest Grok is fundamentally superior across the board, they do confirm that xAI can deliver high-quality, specialized performance when it chooses to focus. For a company still building its reputation, this is a meaningful milestone—and a clear sign that Grok is evolving into a capable, if niche, assistant for gamers and enthusiasts alike.