xAI introduces 1.5-trillion-parameter Grok 4.5 for coding and agent tasks
xAI unveiled Grok 4.5 on July 8, 2026. The mixture-of-experts model trained on Cursor interaction data and scored 83.3% on Terminal-Bench 2.1. It reportedly used about a quarter as many output tokens as Claude Opus 4.8 on comparable tasks. It costs $2/$6 per million input/output tokens, offers three levels of reasoning effort and is available in Grok Build and across Cursor plans.
Visit original source