Introduction to xAI’s Grok 3 Model

Despite the recent countersuit by OpenAI against billionaire Elon Musk, his AI company, xAI, is proceeding with making its flagship Grok 3 model available through an API. This development comes after several months since the unveiling of Grok 3, a model designed to compete with the likes of OpenAI’s GPT-4o and Google’s Gemini.

Features and Capabilities of Grok 3

Grok 3 is notable for its ability to analyze images and respond to questions, powering various features on Musk’s social network, X, which acquired xAI in March. The model is now being offered in two versions through the API: the standard Grok 3 and a smaller version, Grok 3 mini, both of which come with “reasoning” capabilities.

Pricing and Comparison

xAI has set the pricing for Grok 3 at $3 per million tokens input into the model and $15 per million tokens generated by the model. The Grok 3 mini is priced at $0.30 per million input tokens and $0.50 per million output tokens. It’s worth noting that faster versions of both models are available but at a significant premium. In comparison to other models in the market, Grok 3’s pricing matches that of Anthropic’s Claude 3.7 Sonnet but exceeds the cost of Google’s Gemini 2.5 Pro, which generally outperforms Grok 3 in popular AI benchmarks.

Limitations and Controversies

Users on the social network X have pointed out that the Grok 3 model available through xAI’s API has a context window limited to 131,072 tokens, or roughly 97,500 words, which is less than the 1 million tokens xAI initially claimed Grok 3 could support. Furthermore, there have been accusations of misleading benchmark reporting, which could impact the model’s perceived performance and value.

The Evolution of Grok Models

When Elon Musk first announced the Grok model, he positioned it as an edgy, unfiltered, and anti-“woke” AI, willing to tackle controversial questions that other models might avoid. While Grok and its successor, Grok 2, delivered on some of this promise by being more open to discussing sensitive topics, they also demonstrated a tendency to hedge on political subjects and avoid crossing certain boundaries.

Political Bias and Neutralization Efforts

Studies have shown that the earlier Grok models leaned to the political left on various issues, contradicting Musk’s initial stance on the model’s neutrality. Musk has attributed this bias to the model’s training data, primarily public web pages, and has vowed to make the Grok model more politically neutral. However, it remains unclear whether xAI has successfully achieved this goal with Grok 3 and what implications this might have for the model’s use and reception.