SpaceXAI announced the release of Grok 4.5, its most advanced model to date, designed for coding, agentic tasks, and knowledge work. The model was trained alongside Cursor and is intended to deliver intelligent and efficient reasoning for real-world engineering tasks. Grok 4.5 is now the default model in Grok Build, where it excels at complex Excel modeling and office automation tasks. It also performs well in PowerPoint and Word, creating detailed diagrams and clear prose. The model is available today in Grok Build, Cursor, and the SpaceXAI console. Users can start with a few lines of code using an API key. Grok 4.5 is served at 80 TPS and offers twice the token efficiency compared to leading models, reducing costs and processing time. The model is also available for free in Grok Build and Cursor for a limited time. Note that Grok 4.5 is not yet available in the EU, with EU availability expected in mid-July.
The model was trained across tens of thousands of NVIDIA GB300 GPUs, with a focus on data filtering, curation, and reinforcement learning for multi-step software engineering tasks. The training stack supports highly asynchronous operations, enabling long-running agentic rollouts. Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens, offering the highest intelligence per unit of time and cost. The model is capable of creating well-designed, end-to-end functional apps with minimal specification, such as a solar system simulation using three.js. It also excels at coding tasks, including Rust and C/C++ challenges and end-to-end app building. The model's token efficiency is demonstrated by resolving tasks with an average of 15,954 output tokens, about 4.2× fewer than Opus 4.8 (max) at 67,020. Grok 4.5 scores #1 on Harvey's Legal Agent Benchmark and performs well on benchmarks like DeepSWE 1.0, Terminal Bench 2.1, and SWE Bench Pro. The model is also capable of building complex diagrams in PowerPoint and writing clear prose in Word. Users can access Grok 4.5 through the SpaceXAI API, with a limited-time free trial available in Grok Build and Cursor.
The model's availability in the EU is expected in mid-July. Grok 4.5 is trained on datasets spanning coding, science, engineering, and math, with a focus on real-world applications. The model's training process includes reinforcement learning with hundreds of thousands of tasks, automated grading, and domain-specific data selection. The result is a model that delivers intelligent results more quickly and at lower costs. The model is also capable of creating well-designed, end-to-end functional apps with minimal specification, such as a solar system simulation using three.js. It is available for free in Grok Build and Cursor for a limited time. The model's pricing is competitive, with $2 per million input tokens and $6 per million output tokens. The model's token efficiency is demonstrated by resolving tasks with an average of 15,954 output tokens, about 4.2× fewer than Opus 4.8 (max) at 67,020. Grok 4.5 is now the default model in Grok Build, where it excels at complex Excel modeling and office automation tasks.
It also performs well in PowerPoint and Word, creating detailed diagrams and clear prose. The model is available today in Grok Build, Cursor, and the SpaceXAI console. Users can start with a few lines of code using an API key. Grok 4.5 is served at 80 TPS and offers twice the token efficiency compared to leading models, reducing costs and processing time. The model is also available for free in Grok Build and Cursor for a limited time. Note that Grok 4.5 is not yet available in the EU, with EU availability expected in mid-July.
Source: xai
