SpaceXAI has officially released Grok 4.5—its smartest, most efficient model yet, purpose-built to dominate coding, agentic workflows, and complex office tasks.
- Developer Powerhouse: Co-trained with Cursor, Grok 4.5 tackles complex, multi-step engineering tasks, achieving a leading 29.0% resolution rate on the SWE Marathon benchmark.
- Blazing Fast & Token-Efficient: Served at 80 TPS (tokens per second), Grok 4.5 resolves SWE Bench Pro tasks using 4.2× fewer output tokens than competitor models like Opus 4.8.
- Built for the Office: As the default engine in Grok Build, it designs native PowerPoint diagrams, writes polished Word documents, and builds fully functional Excel financial models complete with web research.
- Disruptive Pricing: Highly cost-effective at $2 per million input and $6 per million output tokens.
(Note: EU availability is expected in mid-July 2026).
Sources: @grok, @SpaceXAI, blog post