Cursor released Grok 4.5 with SpaceXAI today. Cursor says the new model is its most intelligent model, and it is a mixture-of-experts model trained jointly with SpaceXAI. For individual and team plan users, the launch comes with double usage for the first week.
Cursor also made Grok 4.5 available across desktop, web, iOS, CLI, and the SDK. That matters because the same model now reaches the places where people write, test, and ship work instead of forcing them into one interface. Cursor is betting that one model can move across software engineering and broader computer-based work without a separate rollout for each surface.
Cursor and SpaceXAI
Cursor said Grok 4.5 is the first model it has built for more than software engineering. It says the model can handle difficult, long-running tasks that require creatively using tools to solve problems. Cursor also says the model can be used in software engineering, data science, finance, legal work, and other computer-based work.
The training mix is part of the story. Cursor said the model was trained on trillions of tokens of Cursor data, including user interactions with codebases and software tools, plus high-quality STEM tasks, research papers, and other knowledge work. Cursor also said it used reinforcement learning on difficult problems in realistic environments spanning software engineering and broader knowledge work.
MoE and Composer 2.5
Cursor added new safeguards reflecting Grok 4.5’s cybersecurity capabilities. That is the piece users will notice only if the model is being asked to operate in more sensitive workflows, where a broader toolset also means more need for guardrails.
The pricing is explicit. Cursor set the base model at $2/M input tokens and $6/M output tokens. The fast variant is priced at $4/M input tokens and $18/M output tokens. Cursor also said Grok 4.5 and Composer 2.5 are different model weight classes, and Composer 2.5 will remain offered while new models of that size are released going forward.
