SpaceX AI unveils Grok 4.5, its next-generation flagship large language model.

SpaceX AI has launched its flagship model, Grok 4.5, which employs a 1.5T parameter MoE architecture and is jointly trained with a Cursor, achieving performance comparable to Claude Opus. The model boasts an inference speed of 80 TPS, surpasses Opus 4.8 in programming capabilities on DeepSWE Bench, and offers over four times the token efficiency. It supports 500,000 contexts, with an upgrade to one million next week, and a 2T parameter version to be released at the end of the month.