SpaceXAI unveils Grok 4.5, Opus-class model aimed at coding and agentic work

SpaceXAI has publicly launched Grok 4.5, its most capable AI model to date and the company’s first major release since going public. The model is positioned as an Opus-class system designed to match or approach Anthropic’s Claude Opus line while offering faster performance, higher token efficiency and lower cost. Grok 4.5 is built on a 1.5 trillion-parameter V9 foundation model and was trained across tens of thousands of Nvidia GB300 GPUs, with SpaceXAI emphasizing heavy investments in data filtering, deduplication and quality scoring to improve reliability and coverage in technical domains.

Functionally, Grok 4.5 targets practical, high-value enterprise workloads, especially coding, finance, legal tasks, and broader agentic and knowledge work. The model was jointly developed with coding agent provider Cursor and leverages trillions of tokens of Cursor interaction data, capturing real-world developer workflows and tool usage. This training approach is intended to make the system particularly strong at long-running, tool-using tasks like working across large codebases, building complex Excel models from web research, and producing structured PowerPoint and office content for professional settings.

From a go-to-market standpoint, Grok 4.5 is immediately available through SpaceXAI’s Grok Build coding agent, inside Cursor across desktop, web, iOS, CLI and SDK, and via the SpaceXAI console for API access. The base pricing is set at $2 per million input tokens and $6 per million output tokens, roughly half the price of many rival frontier models, and the company highlights about 2x or better token efficiency versus comparable systems. Independent agentic benchmarks such as AutomationBench-AA and GDPval-AA v2 report that Grok 4.5 sits on the performance–cost Pareto frontier, with top-tier agentic scores at substantially lower per-task cost than leading competitors.

In terms of availability, Grok 4.5 is live today for users in the United States and other supported regions through the SpaceXAI console, Grok Build, and Cursor, but it is not yet accessible in the European Union. SpaceXAI’s documentation and multiple news reports state that EU access is planned for mid- to late July, following the initial global rollout. The launch marks a strategic push by SpaceXAI toward the enterprise AI market, with a focus on software engineering, legal and financial workflows, and office automation, where lower cost per token and strong agentic performance could be particularly attractive to businesses seeking to scale AI usage.