SpaceXAI launches Grok 4.5 AI model with low‑cost coding and agentic capabilities

SpaceXAI (formerly xAI) released Grok 4.5, a large‑language model aimed at software‑development and autonomous‑agent tasks. The model is priced at $2 per million input tokens and $6 per million output tokens, positioning it as a cheaper alternative to Anthropic’s Opus and OpenAI’s premium offerings. SpaceXAI describes Grok 4.5 as “Opus‑class” in capability, but faster and twice as token‑efficient, processing roughly 80 tokens per second.

The model was trained on the Colossus super‑computer cluster using tens of thousands of Nvidia GB300 GPUs and data from the recently acquired AI‑coding platform Cursor, which supplied real developer‑session logs. Benchmarks show Grok 4.5 scoring 62 % on the DeepSWE 1.0 coding test and leading the SWE Marathon benchmark with a 29 % success rate, while using far fewer tokens per task than competing models.

Within hours of its public launch, the system was breached by hackers, exposing a vulnerability that led to a complete collapse of the model’s defenses. The breach, reported by several outlets, raised immediate security concerns. The same day, SpaceX’s stock slipped 0.8 % to a new low after the announcement.

Overall, Grok 4.5 is presented as a cost‑focused, high‑throughput AI solution for enterprises that need extensive code analysis, multi‑step tool use, and knowledge‑work automation.