SpaceXAI debuts Grok 4.5 aimed at coding and AI agents

SpaceXAI has unveiled Grok 4.5, its most advanced model yet, targeting developers working on code generation and autonomous “agentic” applications.

Announced Wednesday, the company said the model was trained using tens of thousands of Nvidia GB300 GPUs. SpaceXAI highlighted its focus on data quality during training, including filtering, deduplication and quality scoring.

The launch was accompanied by an ecosystem push. Popular AI coding agent Cursor said it has partnered with SpaceXAI to train Grok 4.5.

Agentic tasks refer to AI systems that can plan and execute multi-step workflows with minimal human input. By optimizing Grok 4.5 for coding, SpaceXAI is aiming directly at tools used by software developers and AI agents that build, test and deploy code.

The move comes amid intensifying competition to build models specialized for real-world developer workflows rather than general chat.