SpaceXAI has officially launched Grok 4.5, a reasoning model built for coding, agent tasks, and professional reasoning. It's the first product since SpaceX acquired AI coding platform Cursor, and was co-trained on trillions of interactions from Cursor's platform.
On the SWE Bench Pro benchmark, the model averages just 16,000 output tokens per task — a 4.2x savings over Claude 4.8 Opus, dramatically cutting development costs. It runs at 80 tokens per second. API pricing is $2 per million input tokens and $6 per million output tokens. Beyond coding, users can do live Excel modeling in Grok Build or generate layouts and charts in Word and PowerPoint.
Developers can try Grok 4.5 in Cursor, Grok Build, and the SpaceXAI console, though EU users will have to wait until mid-July.