SpaceXAI (formerly xAI until it was rebranded by SpaceX) has rolled out Grok 4.5 to the public, a new AI model built specifically for coding and engineering work. Grok 4.5 runs on the company’s new V9 foundation, which packs 1.5 trillion parameters, making it about three times larger than the v8-small setup behind Grok 4.3. To train the model, SpaceXAI used its Colossus supercluster in Tennessee running NVIDIA’s current flagship Blackwell chips. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, undercutting Anthropic’s Claude Opus 4.8, which runs $5 per million input tokens and $25 per million output tokens. On the output side, Grok 4.5 comes in at roughly a quarter of the price. On benchmarks, Grok 4.5 scored 64.7 per cent on SWE-Bench Pro, and on Cursor’s CursorBench platform, it handles automated coding tasks for $1.51 each. Elon Musk noted that Grok 4.5 isn’t yet using special software the company is still building; once that software is ready and fine-tuned for NVIDIA’s latest AI chips (the GB300), Musk says the model could run twice as fast or better. The model is ready to use through Cursor, the Grok