Their smartest model just got faster and cheaper.
SpaceXAI has unleashed Grok 4.5, its most intelligent model yet, purpose-built for coding, agentic workflows, and knowledge work.
Trained alongside Cursor across tens of thousands of NVIDIA GB300 GPUs, this powerhouse excels at real-world engineering tasks, outperforming rivals like GPT 5.5 and Opus 4.8 across multiple benchmarks including DeepSWE and Terminal Bench.
The model’s secret weapon? Token efficiency, resolving SWE Bench Pro tasks with 4.2 times fewer outputs than competitors, slashing costs while maintaining blistering 80 TPS speeds.
Beyond coding, Grok 4.5 now powers Grok Build, crafting complex Excel models, polished PowerPoint presentations, and clear Word documents with native shapes and intuitive layouts.
Priced at $2 per million input tokens and $6 per million output, it delivers unmatched intelligence per dollar.
Available today in Grok Build, Cursor, and via API, the model can build full-stack apps from a single prompt, from solar system simulations to enterprise dashboards.
EU users will gain access mid-July.
