SpaceXAI launches Grok 4.5 for coding and agentic tasks
The new Grok model delivers enhanced coding performance, workplace automation tools, and competitive pricing for developers and businesses.
SpaceXAI has introduced Grok 4.5, its latest model for coding, agentic tasks and knowledge work.
The company describes Grok 4.5 as its most capable model to date, trained for real-world engineering, software development, science and mathematics tasks.
According to SpaceXAI, the model was trained across tens of thousands of NVIDIA GB300 GPUs, using data filtering, curation and reinforcement learning focused on multi-step technical work.
The company says Grok 4.5 can complete complex coding tasks and build functional applications from a single prompt.
Beyond software development, Grok 4.5 is designed to support knowledge-work tasks in Grok Build, including Excel modelling, PowerPoint slide design and Word document drafting.
SpaceXAI said the model serves up to 80 tokens per second and offers greater token efficiency than comparable leading models on selected tasks.
Grok 4.5 is available through Grok Build, Cursor and the SpaceXAI console. It is priced at $2 per million input tokens and $6 per million output tokens.
The company said Grok 4.5 is not yet available in the EU through its products or API console, with the EU availability expected in mid-July.
Why does it matter?
Grok 4.5 shows how frontier AI competition is moving towards practical agentic work, not only general chatbot performance. Coding, spreadsheet modelling, document creation and application-building are becoming key battlegrounds for AI providers targeting enterprise and professional users. The model’s pricing and claimed token efficiency also show growing pressure on AI companies to compete on cost and speed as much as benchmark performance.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!