Elon Musk’s XAI Releases Grok 3 API
Although Elon Musk is now facing a lawsuit from OpenAI, his AI venture xAI continues to push forward. Most recently, it introduced API access for its latest model, Grok 3, signalling a bold move in the competitive AI space. This comes several months after Grok 3 was first unveiled as a direct challenger to OpenAI’s GPT-4o and Google’s Gemini.
Grok 3 can interpret images, respond to complex queries, and powers several features on X, Musk’s social platform. Interestingly, xAI itself was acquired by X earlier this year. Through its API, xAI offers two model versions: Grok 3 and a lighter Grok 3 Mini, both featuring “reasoning” capabilities.
When it comes to pricing, xAI hasn’t held back. Grok 3 costs $3 per million input tokens and $15 per million output tokens. By contrast, Grok 3 Mini is significantly cheaper, priced at $0.30 and $0.50 per million tokens for input and output respectively. For users seeking faster performance, premium variants are available at even higher rates — with Grok 3 peaking at $25 per million output tokens.
Competitive Pricing and Capabilities Raise Questions
Despite the features on offer, Grok 3 isn’t considered a bargain. Its pricing structure closely mirrors that of Anthropic’s Claude 3.7 Sonnet, another reasoning-enabled model. However, Grok 3 lags behind Google’s Gemini 2.5 Pro in benchmark tests, which adds to questions about its value. Adding to this, xAI has faced criticism for allegedly misrepresenting Grok 3’s performance in benchmarks.
There’s also a notable discrepancy in context window capacity. While xAI previously claimed Grok 3 could handle up to one million tokens, the current API version caps out at just over 131,000 tokens — a significant shortfall. This limitation could affect how developers adopt the tool for large-scale applications.
The Politics and Personality of Grok
Originally pitched by Musk as bold, uncensored, and anti-“woke,” earlier Grok models lived up to their rebellious branding — particularly when prompted to be crude or provocative. Nonetheless, when it came to politics, those earlier versions held back. One academic study even found Grok leaned left on issues like diversity, inequality, and transgender rights.
Musk attributes that behaviour to Grok’s public web training data and has promised to steer the model toward neutrality. While efforts are reportedly underway, it remains uncertain whether these adjustments have made a measurable difference. Aside from a few controversial missteps — such as censoring criticisms of Musk and Donald Trump — Grok 3’s ideological stance is still being scrutinised. The long-term effects of this editorial tweaking are yet to be fully understood.
