Despite an ongoing legal battle with OpenAI, Elon Musk’s xAI has taken another significant step forward by launching a public API for Grok 3, its flagship artificial intelligence model. Grok 3 was first unveiled several months ago as the company’s direct competitor to prominent AI systems such as OpenAI’s GPT-4o and Google’s Gemini. Through capabilities like image analysis and question answering, Grok 3 has already been integrated into select features of Musk’s social media platform, X, which notably acquired xAI back in March.
The Grok 3 API is available in two primary versions: the standard Grok 3 and a scaled-down variant known as Grok 3 Mini, optimized to deliver enhanced reasoning functions. Pricing details reveal Grok 3 as a relatively premium offering, costing developers $3 per million input tokens (around 750,000 words) and $15 for every million output tokens generated by the model. By contrast, Grok 3 Mini is priced significantly lower, at just $0.30 per million input tokens and $0.50 per million output tokens. Additionally, faster-performance variations of these two models are provided at higher price points—$5 per million input tokens and $25 per million generated tokens for Grok 3, with Grok 3 Mini’s premium tier priced at $0.60 per million tokens input and $4 per million output.
Notably, the Grok 3 API is priced similarly to Anthropic’s advanced Claude 3.7 Sonnet model, which also emphasizes reasoning capacity. However, it is more costly than Google’s newest Gemini 2.5 Pro, a model that typically achieves superior scores compared to Grok 3 across established AI assessment benchmarks. Earlier this year, xAI came under scrutiny for alleged misrepresentation of Grok 3’s benchmark performance data.
Additionally, some users have raised concerns over Grok 3’s advertised capabilities. While xAI initially claimed Grok 3 would support a context window of up to one million tokens for processing extensive inputs, the API is currently limited to a significantly reduced 131,072-token limit—about 97,500 words.
Since his initial announcement nearly two years ago, Elon Musk promoted Grok as a uniquely outspoken and uncensored AI model, positioned explicitly against the perceived cautiousness and moderation observed in other large language models. Early versions of Grok gained notoriety for easily delivering vulgar or provocative responses. However, studies revealed that earlier Grok iterations would nevertheless shy away from sensitive political topics and leaned politically leftward on issues such as transgender rights and equality programs.
Musk attributed these biases to Grok’s training data, sourced largely from publicly accessible web pages, and vowed publicly to steer the AI’s development towards political neutrality. It remains unclear how fully xAI has realized this ambition or if Grok 3 has successfully addressed previous complaints about bias and controversial responses.
