The AI industry barely has time to catch its breath between major launches. SpaceXAI is the latest to make its move, introducing Grok 4.5 and claiming its newest model outperforms rivals on the tasks developers care about most, including coding and complex reasoning.
According to the company, Grok 4.5 was trained on massive datasets covering software development, science, engineering, and mathematics. It was also developed alongside Cursor, the popular AI coding editor, with the goal of helping developers write software, fix bugs, and tackle demanding technical work more effectively.
Of course, no AI launch would be complete without benchmark charts. SpaceXAI says Grok 4.5 scored 66.1 percent on the DeepSWE benchmark, edging out GPT 5.5 xhigh, which posted 64.31 percent. Those numbers certainly look impressive, but benchmarks rarely tell the whole story. Developers usually care more about whether an AI actually saves time on real projects than how it performs on carefully selected tests.
SpaceXAI also says Grok 4.5 is more efficient than competing models, generating roughly half as many tokens on software engineering tasks while delivering responses at speeds of up to 80 tokens per second. If those claims hold up outside polished demos, the combination of faster responses and lower token usage could translate into meaningful savings for businesses using the API.
The company showcased Grok 4.5 building complete web applications from a single prompt, creating sophisticated Excel workbooks that incorporate web research and multi-sheet formulas, designing PowerPoint presentations with native shapes, and writing polished Word documents. Those demonstrations are certainly impressive, although launch demos tend to represent best-case scenarios.
Pricing is another area where SpaceXAI is trying to stand out. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. The company also claims the model is about twice as token-efficient as comparable rivals, potentially reducing overall costs even further.
Grok 4.5 is available now through Grok Build, Cursor, and the SpaceXAI API. SpaceXAI is also offering free access for a limited time through Grok Build and Cursor. The only notable limitation is that the model is not yet available in the European Union, although availability is expected later this month.
There is certainly no shortage of powerful AI models these days. OpenAI, Anthropic, Google, and now SpaceXAI are all competing to win over developers with increasingly capable models. Whether Grok 4.5 truly delivers on SpaceXAI’s claims will become much clearer once developers start putting it through its paces in real-world projects instead of benchmark tests.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz