Grok 4.5: new AI model has improvements in programming, more speed and lower cost | SpaceXAI | Elon Musk | Anthropic

SpaceXAI has unveiled its new updated Grok 4.5 artificial intelligence (AI) model, specifically designed to execute agent and scheduling tasks with “leading speeds” and greater cost efficiency than its competitors, while maintaining capabilities on par with Anthropic’s Opus 4.7 model.

The company led by Elon Musk has defined its new model as “the most intelligent” that, trained with the Cursor code editor, offers improvements in programming, coding and high capabilities for “engineering in the real world.”

This includes tasks such as creating complex and functional applications even with minimum specifications, as well as tasks related to office work, such as creating complex Excel models, designing content for slides in PowerPoint or writing texts in Word, as well as knowledge-based research work and agent tasks.

It should be noted that Grok 4.5 is capable of executing all these tasks “faster than ‘flash’ models”, as the company has stated in a statement on its blog, where it has indicated that its new model offers a processing speed of 80 transactions per second (TPS).

Likewise, it has been trained optimizing reinforcement learning, with data sets that cover programming, science, engineering and mathematics knowledge. As a result, the company claims that it outperforms comparable models in the aforementioned tasks.

Although it does not reach other leading models such as Fable from Anthropic or GPT 5.5 from OpenAI, Elon Musk himself defines Grok 4.5 as an Opus class model – also Anthropic – but “faster, more efficient in tokens and at a lower cost”, as he shared in a publication on the social network X.

Specifically, he has compared it with the Opus 4.7 model, although it achieves a “more competitive” result, by combining capacity with speed and cost, for real applications. “We are closing the circle in terms of usefulness in the real world, not in ‘benchmarks’,” he stated in another publication.

Regarding its cost efficiency, Musk has also reiterated its “competitive” price compared to other models in the sector. Specifically, it costs $2 per million input tokens and $6 per million output tokens.

With this, it achieves a token efficiency “twice greater” than that of the most recent models in the same range. This translates into a model that provides “intelligent results faster and at a much lower cost.”

Grok 4.5 is now available to users through SpaceXAI, Grok Build and Cursor, however, it has not yet been released in the European Union, where it will arrive “later this month”, as the company has noted.

By Editor