Billionaire Elon Musk has just received a counter-suit from OpenAI, but that hasn’t stopped his company xAI from making its flagship Grok 3 model available via API.
It’s been a few months since xAI introduced the Grok 3, the company’s answer to models like OpenAI’s GPT-4o and Google’s Gemini. Grok 3 can analyze images and answer questions, and powers a number of features on Elon Musk’s X social network, which was not coincidentally acquired by xAI in March.
Through its API, xAI offers two versions of its flagship model: Grok 3 and Grok 3 Mini with “reasoning” capabilities.
Grok 3 is priced at $3 per million tokens (~750,000 words) input to the model and $15 per million tokens generated by the model. Meanwhile, Grok 3 Mini will cost $0.30 per million input tokens and $0.50 per million output tokens.
Faster versions of Grok 3 and Grok 3 Mini are available for an additional fee: $5 per million incoming tokens and $25 per million outgoing tokens for Grok 3; and $0.60 per million incoming tokens and $4 per million outgoing tokens for Grok 3 Mini.
Grok 3 is not cheap compared to its competitors. xAI is in line with the price of Anthropic’s Claude 3.7 Sonnet, which also offers reasoning capabilities, but it is more expensive than Google’s recently released Gemini 2.5 Pro, which achieves generally higher results than Grok 3 on popular AI tests. (Notably, xAI has been accused of being misleading in its reports of Grok 3’s test results.)
As several users on X have pointed out, Grok 3, via the xAI API, also has a smaller context window than the model is supposedly capable of supporting. (“Context window” refers to how many tokens a model can process in a single pass.) The API supports a maximum of 131,072 tokens, or roughly 97,500 words – less than the 1 million tokens that xAI claimed to support in late February with Grok 3.
When Musk announced Grok about two years ago, he presented the AI model as sharp, unfiltered, and anti-“woke” – in short, as ready to answer controversial questions that other AI systems fail to answer. To some extent, it has kept this promise. For example, if you ask Grok and Grok 2 to be vulgar, they will gladly do so, spewing colorful language that you probably won’t hear in ChatGPT.
But the Grok models prior to Grok 3 avoided political topics and did not cross certain boundaries. In fact, one study found that Grok leaned toward left-wing political views on topics such as transgender rights, diversity programs, and inequality.
Musk blamed this behavior on Grok’s training data-public web pages-and promised to “move Grok closer to a politically neutral stance.” Aside from high-profile mistakes like briefly censoring unflattering references to President Donald Trump and Musk, it is unclear whether xAI has achieved this goal at the model level and what the long-term consequences might be.
