AI BriefingGrok 4.5 Launched
Key point
SpaceXAI has released Grok 4.5, its latest model optimized for coding, agentic tasks, and knowledge work.
Details
SpaceXAI has unveiled Grok 4.5, its most powerful model yet, delivering outstanding performance in coding, agentic tasks, and knowledge work. This model was trained together with Cursor, and was built on a massive dataset spanning coding, science, engineering, and mathematics.
Grok 4.5 demonstrates performance that surpasses existing leading models in real-world engineering tasks. Its key features are as follows.
- Engineering and Agentic Performance: It recorded industry-leading performance on major benchmarks including DeepSWE, Terminal Bench, and SWE Bench Pro.
- Overwhelming Efficiency: It delivers fast inference at 80 TPS, with token efficiency roughly 4.2x higher than existing models for the same task.
- Powerful Training Approach: It was trained using tens of thousands of NVIDIA GB300 GPUs, maximizing reasoning capability through data filtering and reinforcement learning (RL).
In addition, it has been adopted as the base model for Grok Build, supporting complex Excel model creation, web research-based data organization, and even sophisticated document work using PowerPoint and Word.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.