SpaceXAI Ships Grok 4.5 With Cursor, EU Access Delayed

The pitch Axios laid out today is straightforward. SpaceXAI shipped Grok 4.5, its first model trained alongside Cursor, priced at $2 per million input tokens and $6 per million output tokens, live now in Grok Build, Cursor on all plans, and the SpaceXAI console. Musk described it as an 'Opus-class' model, claiming faster performance and lower operating costs than the tier it is being compared to.

The angle worth pausing on is who this is being sold to. The company's framing is that Grok 4.5 can handle 'difficult, long-running tasks' in software engineering, data science, finance and legal work, exactly the white-collar knowledge work Anthropic and OpenAI have been staking out. On its own product page, xAI claims the model performs on par with GPT 5.5 and close to Opus 4.8 on coding benchmarks and scores #1 on Harvey's Legal Agent Benchmark. Those are company-reported numbers rather than independent evaluations, so take the specifics as reported, not settled.

The Cursor collaboration is the more structurally interesting piece. Training a model alongside an IDE vendor means access to real coding-session behavior during development, and it hands xAI a distribution channel already attached. Grok 4.5 is available inside Cursor on every plan at launch, which is a very different go-to-market from an API-only release, and it puts a frontier-tier xAI model in front of working developers on day one.

The honest caveat is what the reporting does not give you. There is no EU availability, with the company saying it is expected in mid-July, and there is no independent confirmation of the benchmark claims or the Opus-class framing. Nothing in the coverage explains the commercial terms of the Cursor arrangement, whether that is revenue share, exclusivity, or how coding-session data flows back into training. Those are the questions a buyer would ask before rewriting an agent pipeline around it.

For teams already paying flagship-tier rates on long-context agentic work, the interesting number is the price line, and Musk's pitch is explicitly that Grok 4.5 is faster and cheaper than the Opus tier it is compared to. If that quality claim survives contact with real workloads, the pricing conversation at the frontier just moved.