Notably, Grok 3 features an innovative “DeepSearch” capability that enhances its reasoning by analyzing information from multiple sources. However, early testing revealed some limitations, including issues with citation accuracy and understanding humor. Subscribers to xAI’s Premium Plus tier will gain early access to Grok 3, with additional features available through a new subscription plan called Super Grok. xAI plans to open-source Grok 2, aiming for a politically neutral approach while navigating past criticisms over biases in its models. Andrej Karpathy shared his experience with Grok 3 on social media, stating that he was among the first to access the latest version of the software. He described Grok 3 as having a state-of-the-art thinking model, particularly praising its performance with the board game Settler’s of Catan.
More power doesn’t automatically mean better AI, as we learned from DeepSeek. xAI is clearly ramping up its training scale. The mention of legal documents hints at a push for credibility in regulated industries (or possibly an attempt to sidestep previous criticisms about misinformation).
Scoring well on math tests doesn’t guarantee usefulness for businesses or consumers. Many AI models ace benchmarks but still struggle in practical applications. xAI has positioned as politically neutral, but AI neutrality is nearly impossible. The open-source move might attract developers, but it also raises concerns about who controls and fine-tunes the model’s outputs.
And those questions of trust become important as the models move to commodity. You can pick any model – why pick this one?