Overview
- SpaceXAI made Grok 4.5 publicly available as a production model built on a V9 foundation with about 1.5 trillion parameters trained on the Colossus supercluster in Tennessee using NVIDIA Blackwell chips.
- The company published list pricing of $2 per million input tokens and $6 per million output tokens to position Grok 4.5 as a substantially cheaper option for heavy coding workloads compared with several rivals.
- SpaceXAI released internal benchmark results—a 64.7% score on SWE‑Bench Pro and roughly $1.51 per automated coding task on CursorBench—while noting many of those measurements come from company‑controlled tests.
- The model is accessible immediately through developer tools including Cursor, the Grok Build CLI, Vercel, the developer API, and linked OpenClaw subscriptions, as SpaceXAI pushes for fast adoption in engineering workflows.
- Elon Musk said forthcoming runtime software optimizations could roughly double Grok 4.5’s speed and cost efficiency, but independent verification of performance, comparative claims, and broader safety or regulatory impacts is still needed.