xAI Releases Grok 4.5
xAI has released Grok 4.5, a new model built for coding and agentic tasks, developed in collaboration with the Cursor team—the editor that moved under SpaceX control in June. On Terminal Bench 2.1, Grok 4.5 nearly matches GPT-5.5 and trails Fable 5 by just one point, though it lags behind leaders on GitHub-based problem solving (DeepSWE 1.1).
The model is priced at $2/1M input tokens and $6/1M output tokens, with a generation speed of up to 80 tokens/second. According to xAI, Grok 4.5 uses 4.2× fewer tokens than Opus 4.8 on SWE Bench Pro. It is available in the xAI console, Grok Build, and the Cursor editor, with official plugins released for Word, PowerPoint, and Excel. The EU release is delayed until mid-July.