With Grok 4.5, SpaceXAI introduces a new general AI model that isn’t better than the competition but is cheaper.
SpaceXAI (until recently just xAI) has launched Grok 4.5. The model is immediately available in Grok Build, in Cursor on all formulas, and via the company’s API console, although the latter option does not yet exist in the EU for the time being.
‘Opus-class’
Grok 4.5 is an Opus-class model, according to big boss Elon Musk himself. It reportedly performs at the level of Opus 4.7. That in itself is not spectacular. Anthropic is now at Opus 4.8 and subsequently launched the much more powerful Fable. OpenAI also delivers models that are more powerful.
SpaceXAI does not intend to compete on pure capabilities with Grok 4.5. According to the company, the new LLM excels in efficiency. SpaceXAI states that Grok delivers the same level of performance as equivalent models, but at half the cost. Grok 4.5 is said to handle tokens twice as efficiently.
Lowest price guarantee
xAI explicitly positions Grok 4.5 as a cheaper alternative to other large models. The price is two dollars per million input tokens and six dollars per million output tokens. Combined with the higher efficiency, this should reduce the total cost per completed task.
If SpaceXAI’s claims are correct, they represent a relevant selling point. For many tasks, LLMs are already powerful enough today. Especially regarding knowledge work, you don’t need Fable to achieve good results. With good data and context, a lighter model can manage just fine. Cost is a very relevant factor in that segment.
Support from big names
In recent weeks, SpaceXAI has also worked its way into the good graces of companies that bring LLMs to customers. For instance, both Snowflake and Databricks announced they would support Grok in their platforms.
Such companies (but also SAP, Salesforce, ServiceNow…) provide the context that is so essential for good results. The model provider doesn’t have to worry about that themselves. An LLM doesn’t need to be smarter and more complex than necessary: smart enough and as efficient as possible is a more interesting combination of traits.
Simple switch
By competing on price, SpaceX can work its way into companies. After all, switching models is no longer that complex for businesses that integrate AI in a well-managed way. Switching LLMs can be done almost at the touch of a button, and a lower price can make that mouse click very attractive.
SpaceXAI must hope that organizations follow that reasoning. The AI department, part of SpaceX, is after all responsible for an unprecedented IPO of more than a trillion dollars. This is based on promises of profit, which at this moment cannot be substantiated with real figures. In the long run, investors will want to see results.
Controversial model
Whether a sharp price tag will be enough to lure organizations to Grok en masse remains to be seen. Under Musk’s leadership, SpaceXAI has profiled itself in recent months as the least ethical of all the AI specialists.
Musk does not like restrictions and guardrails, and that has exactly the consequences you would expect. For example, earlier versions of Grok easily created sexually explicit deepfakes, including those of celebrities. Grok was also not shy about some racism and anti-Semitism in the past. In 2025, the latest version of the LLM began calling itself ‘MechaHitler’.
The fact that Grok 4.5 is not available in the EU at launch, while AI models from OpenAI and Anthropic usually cross the Atlantic immediately, is not a sign of confidence in the model’s ability to actually follow the rules properly.
Furthermore, SpaceX performs particularly poorly on almost all ESG rankings, partly thanks to the controversial figure of Musk himself. That is enough to keep the company out of most investment funds in Europe, and may also play a role in the decision of companies to embrace Grok for their AI needs.
