Grok 3 API Released by xAI - Elon Musk's AI Company

Elon Musk's xAI Launches Grok 3 API Amidst Legal Dispute

Despite a recent countersuit filed by OpenAI, Elon Musk’s artificial intelligence firm, xAI, is now providing access to its primary language model, Grok 3, through an Application Programming Interface (API).

Grok 3: A New Contender in the AI Landscape

Several months have passed since xAI initially presented Grok 3, positioned as a competitor to leading models such as OpenAI’s GPT-4o and Google’s Gemini. This model possesses the ability to interpret images and formulate responses to queries. It currently underpins several functionalities within Musk’s social media platform, X, which xAI acquired in March.

API Offerings: Grok 3 and Grok 3 Mini

xAI is making two versions of its core model available via its API: the full Grok 3 and a more streamlined Grok 3 Mini, both equipped with reasoning capabilities.

The pricing structure for Grok 3 is set at $3 for every million tokens inputted (~750,000 words) and $15 per million tokens generated. Grok 3 Mini is offered at a lower cost of $0.30 per million input tokens and $0.50 per million output tokens.

Premium Speed Options

Faster processing speeds for both Grok 3 and Grok 3 Mini are available at a higher price point. Grok 3’s expedited version costs $5 per million input tokens and $25 per million output tokens. The faster Grok 3 Mini is priced at $0.60 per million input tokens and $4 per million output tokens.

Cost Comparison and Benchmarks

Grok 3’s pricing aligns with Anthropic’s Claude 3.7 Sonnet, which also features reasoning abilities. However, it is more expensive than Google’s recently released Gemini 2.5 Pro, which generally achieves superior results than Grok 3 on widely used AI performance evaluations. (It is worth noting that xAI has faced accusations of presenting potentially misleading information in its Grok 3 benchmark reports.)

Context Window Limitations

Observations from users on X have revealed that the Grok 3 API currently supports a smaller context window than initially advertised. The “context window” defines the amount of tokens the model can process simultaneously. The API is currently limited to 131,072 tokens, equivalent to approximately 97,500 words – falling short of the 1 million token capacity xAI stated Grok 3 would support as of late February.

Grok's Original Positioning and Evolution

Upon its initial announcement roughly two years ago, Musk characterized Grok as a bold, uncensored, and anti-"woke" AI model – designed to address controversial questions that other AI systems might avoid. To a degree, this promise was fulfilled.

For instance, when prompted to generate vulgar content, both Grok and Grok 2 readily complied, producing language unlikely to be encountered from ChatGPT.

Political Nuances and Training Data

However, earlier Grok models exhibited hesitancy on politically charged topics and avoided crossing specific boundaries. In fact, a study indicated that Grok demonstrated a leaning towards the political left on issues such as transgender rights, diversity initiatives, and economic inequality.

Musk attributed this behavior to the model’s training data – sourced from public web pages – and committed to adjusting Grok towards a more politically neutral stance. Aside from isolated incidents, such as the temporary censoring of unfavorable references to President Donald Trump and Musk himself, it remains uncertain whether xAI has successfully achieved this neutrality at the model level and what the long-term implications may be.