SpaceXAI and Cursor ship Grok 4.5
Grok 4.5 is the first model built with Cursor since the $60B acquisition agreement, targeting enterprise work in legal, finance, and software engineering.
SpaceXAI and Cursor jointly released Grok 4.5 on July 8, the first model built with Cursor’s infrastructure since SpaceX signed a $60 billion all-stock acquisition agreement with Anysphere (Cursor’s parent company) in June 2026. The deal has not yet closed; closing is expected in Q3 2026. Grok 4.5 ships in Cursor on all plans, on Grok Build, and on the SpaceXAI API console. EU availability is delayed to mid-July; no reason was given.
Grok 4.5 is not a general chatbot. It is designed for extended, multi-step knowledge work: software engineering, legal work, financial analysis, data science, and cybersecurity are the named use cases.
Model details
Grok 4.5 has 1.5 trillion parameters and is described internally as a V9 model. Pre-training incorporated Cursor workflow data in a supplemental stage. Post-training ran on Colossus, SpaceXAI’s in-house supercomputer housing 555,000 NVIDIA GB300 GPUs at an approximate cost of $18 billion.
Standard pricing is $2 per million input tokens and $6 per million output tokens. A premium version costs $4 per million input and $18 per million output. For comparison, Anthropic’s Opus 4.8 is priced at $5 input and $25 output per million tokens at list price.
Company-reported benchmarks
SpaceXAI reports Grok 4.5 ranks first on Harvey’s Legal Agent Benchmark, performs on par with GPT-5.5 on coding evaluations, and scores close to Opus 4.8 on coding tasks. Musk has stated it beats Opus 4.8 on “several benchmarks.” These figures are company-reported and have not been independently verified as of publication.
What this means for Cursor users
The acquisition agreement changes the strategic context for Cursor users in a material way: the underlying model provider and the IDE are now controlled by the same parent company. For now, Cursor continues to support other model providers, and the acquisition has not yet closed.
For operators choosing between models in Cursor, Grok 4.5’s pricing undercuts Opus 4.8 substantially. The relevant question is whether quality on your specific task holds. Running a side-by-side on a representative sample of your highest-volume coding or analysis tasks is worth the time before switching.
The guide on coding agents for non-developers covers how to evaluate these tools without a benchmarking setup. On the competitive model landscape, Claude Code’s shift to manual mode default from last week shows how the major coding tools are all adjusting their defaults in parallel.