Elon Musk's xAI launches Grok 3 API

Elon Musk might be busy trading lawsuits with OpenAI, but his AI venture is still pushing code. xAI has officially rolled out API access for its flagship Grok 3 model, bringing its latest frontier intelligence to third-party developers.

It has been a few months since Grok 3 was first teased as Musk's direct answer to GPT-4o and Google's Gemini. The multimodal model, which handles text and vision, already underpins several native features across X, which xAI officially acquired back in March.

Matching Claude's rates, with a speed tax

xAI is offering two primary flavors through the endpoint: the full Grok 3 flagship and Grok 3 Mini, which packs dedicated reasoning chops.

Standard Grok 3 is priced at $3 per million input tokens and $15 per million output tokens. For context, one million tokens equates to roughly 750,000 words.

Grok 3 Mini is significantly lighter on the wallet, clocking in at $0.30 per million input tokens and $0.50 per million output tokens.

Interestingly, xAI is also introducing turbo tiers for teams that need faster latency. Speedier Grok 3 jumps to $5 per million input and $25 per million output tokens, while fast Grok 3 Mini runs $0.60 input and $4 output.

A shrinking context window and fierce rivals

Despite the hype, Grok 3 enters the developer arena with some real headwinds.

Its standard pricing mirrors Anthropic's Claude 3.7 Sonnet dollar for dollar. However, it sits notably higher than Google's Gemini 2.5 Pro, which generally outscores Grok 3 across major industry benchmarks. It does not help that xAI's initial benchmark reporting drew scrutiny for questionable comparisons.

Then there is the context size. Several developers on X quickly caught that the API caps out at 131,072 tokens, or about 97,500 words. That is a noticeable backtrack from the 1 million token window xAI boasted about back in late February.

There is also the question of personality. Musk originally promised an unfiltered, anti-woke engine willing to tackle third-rail topics. While Grok has no problem dropping profanities when prompted, earlier studies revealed the model actually leaned left on social issues, and it previously suffered embarrassing hiccups like temporarily silencing critical mentions of Musk and Donald Trump.

Why it matters

For everyday developers, Grok 3 is a tough sell at launch. Unless you specifically need its quirky persona or are building bespoke automation inside the X ecosystem, Claude 3.7 Sonnet remains the more reliable reasoning partner at the exact same price point, while Gemini 2.5 Pro beats it on raw benchmark value.

For the AI landscape, this launch turns xAI into an open enterprise combatant rather than just an in-house playground. Musk can no longer rely purely on personal hype. Models living in an API have to justify their cost-per-token against hardened infrastructure from Google, Microsoft, and Amazon.

In our view, launching with a pruned 131k context window while matching top-tier pricing feels premature. Grok 3 has plenty of raw horsepower, but asking premium rates without delivering on the promised 1-million-token capacity looks like a tactical misstep. Unless xAI rolls out aggressive enterprise discounts, mainstream engineering teams will likely stick with what they already know.