In surprising events, billionaire Elon Musk finds himself countersued by OpenAI. Despite the legal challenges, his AI venture, xAI, is pressing forward, announcing the availability of its flagship Grok 3 model through an API.
Several months have passed since xAI introduced Grok 3, positioning it as the company’s response to models such as OpenAI’s GPT-4o and Google’s Gemini. Grok 3 can analyze images and respond to inquiries, serving as the driving force behind various features on Musk’s social network, X. This development follows the platform’s acquisition by xAI in March.
xAI has introduced two versions of its flagship model through its API: Grok 3 and Grok 3 Mini, featuring enhanced reasoning capabilities.
The pricing structure for Grok 3 is set at $3 for every million tokens, which equates to approximately 750,000 words input into the model. Additionally, the cost for tokens produced by the model stands at $15 per million tokens. In a recent announcement, it has been revealed that Grok 3 Mini will be priced at $0.30 for every million input tokens and $0.50 for each million output tokens.
Enhanced iterations of Grok 3 and Grok 3 Mini are now offered at a premium price point: $5 for every million input tokens and $25 for each million output tokens for Grok 3, while Grok 3 Mini is priced at $0.60 per million input tokens and $4 per million output tokens.
Grok 3’s price tag sets it apart from its competitors. xAI is aligning its pricing with that of Anthropic’s Claude 3.7 Sonnet, a model that provides reasoning capabilities. However, it comes at a higher cost than Google’s newly launched Gemini 2.5 Pro, which consistently outperforms Grok 3 in various widely recognised AI benchmarks. xAI faces allegations of misleading practices concerning its Grok 3 benchmark reports.
Multiple users on X have highlighted that Grok 3, accessed through xAI’s API, features a context window notably smaller than what the model is reportedly designed to accommodate. The term “context window” pertains to the number of tokens the model can process simultaneously. The API has a maximum capacity of 131,072 tokens, translating to approximately 97,500 words. This figure falls short of the 1 million tokens that xAI asserted Grok 3 could handle in late February.
Two years ago, Musk introduced Grok, presenting the AI model as bold, unrestrained, and resistant to mainstream ideologies—essentially, prepared to tackle contentious inquiries that other AI systems shy away from. He fulfilled certain aspects of that commitment. In a bold departure from conventional norms, Grok and Grok 2 are unreserved in their use of language, delivering a barrage of colorful expressions that starkly contrast with the more restrained dialogue typically associated with ChatGPT.
However, earlier versions of Grok models were cautious regarding political topics and refrained from crossing specific lines. A recent study revealed that Grok tended to the political left regarding issues such as transgender rights, diversity initiatives, and inequality.
Musk has attributed that behavior to Grok’s training data, which consists of public web pages, and has committed to making Grok more politically neutral. While there have been notable missteps, such as the temporary censorship of unflattering references to President Donald Trump and Elon Musk, it remains uncertain if xAI has successfully met its objectives at the model level and the potential long-term implications of these actions.