Even though OpenAI just countersued billionaire Elon Musk, his AI startup xAI is still releasing its flagship Grok 3 model through an API.
Grok 3, xAI’s response to models like as Google’s Gemini and OpenAI’s GPT-4o, was released a few months ago. Grok 3, which drives several features of Musk’s social network X—which, incidentally, was acquired by xAI in March—is capable of image analysis and question answering.
Grok 3 and Grok 3 Mini with “reasoning” features are the two versions of xAI’s flagship model that are available through its API.
Grok 3 costs $3 for each million tokens (around 750,000 words) that are fed into the model and $15 for each million tokens that the model produces. Grok 3 Mini, on the other hand, will cost $0.50 per million output tokens and $0.30 per million input tokens.
Grok 3 Mini and Grok 3 are faster variants, although they cost more: Grok 3 Mini costs $0.60 per million input tokens and $4 per million output tokens, while Grok 3 costs $5 per million input tokens and $25 per million output tokens.
In comparison to the competitors, Grok 3 is not inexpensive. Although xAI is priced similarly to Anthropic’s Claude 3.7 Sonnet, which also has reasoning skills, it costs more than Google’s recently announced Gemini 2.5 Pro, which outperforms Grok 3 on a number of well-known AI benchmarks. (Notably, xAI’s Grok 3 benchmark reports have been criticized for being deceptive.)
Grok 3 via xAI’s API also has a smaller context window than the model is allegedly able to support, as multiple users on X noted. “Context window” describes the maximum number of tokens the model can process at once. The maximum amount of tokens that the API can handle is 131,072 (around 97,500 words), which is less than the 1 million tokens that xAI stated Grok 3 could manage in late February.
Musk marketed Grok as edgy, uncensored, and anti-“woke” when he first unveiled it about two years ago. In other words, the AI model is generally willing to respond to contentious queries that other AI systems won’t. He somewhat fulfilled that pledge. For instance, Grok and Grok 2 would gladly comply with requests to be profane, using colorful language that you probably wouldn’t hear from ChatGPT.
However, Grok models until Grok 3 avoided political topics and stayed within defined bounds. According to one survey, Grok actually leans left on issues like inequality, diversity initiatives, and transgender rights.
Musk has blamed that conduct on Grok’s training data — public web pages — and committed to “shift Grok closer to politically neutral.” It’s unclear at this time if xAI has succeeded in achieving that goal at the model level and what the long-term ramifications would be, aside from high-profile errors like momentarily filtering disparaging remarks about Musk and President Donald Trump.