SpaceXAI has released Grok 4.5, its most capable AI model to date, purpose-built for coding, agentic workflows, and complex knowledge work. The launch marks the company’s first major model release following its public listing and the acquisition of the AI-powered code editor Cursor. Grok 4.5 stands out for its focus on practical engineering challenges rather than general conversation.
It handles demanding software development tasks, from intricate low-level programming in languages like Rust and C++ to building complete applications from simple prompts.
Developers can generate functional, production-ready apps with minimal instructions, including interactive simulations with realistic physics and polished user interfaces.
The model’s training emphasized real-world performance. It draws from extensive datasets in coding, science, engineering, and mathematics, refined through careful filtering, deduplication, and quality assessment.
A significant portion of its development involved close collaboration with Cursor, incorporating insights from actual developer sessions and workflows.
This supplemental training on Cursor data helped enhance its ability to manage multi-step agentic tasks, long-running processes, and iterative problem-solving typical in professional software engineering.
Performance benchmarks highlight its strengths in coding and terminal-based environments.
On key evaluations such as SWE Bench variants and Terminal Bench, Grok 4.5 delivers competitive or superior results compared to leading alternatives, while requiring far fewer output tokens to complete tasks.
It runs at high inference speeds around 80 tokens per second, combining strong reasoning with efficiency that reduces both time and cost for users.
Beyond pure coding, Grok 4.5 excels in office productivity within the Grok Build environment.
It can construct sophisticated Excel models involving web research and multi-sheet formulas, create detailed PowerPoint presentations with native shapes and diagrams, and draft clear documents in Word.
These capabilities make it suitable for technical professionals handling both development and analytical work.
Pricing reflects the model’s efficiency focus. It is offered at $2 per million input tokens and $6 per million output tokens through the SpaceXAI API—significantly more affordable than many frontier competitors while delivering high intelligence per token and per dollar spent.
Grok 4.5 is now available immediately as the default model in Grok Build, across all Cursor subscription plans, and via the SpaceXAI developer console with an API key.
It is not yet accessible in the European Union, though rollout there is planned for mid-July.
This release underscores SpaceXAI’s strategy of integrating advanced models directly into developer tools through its Cursor partnership. By combining frontier-scale training with targeted coding data and efficient serving, Grok 4.5 aims to accelerate real engineering output while lowering barriers for teams working on complex projects.
