xAI Releases "Grok 4.5" — Fast and Low-Cost, Built for Coding and Agentic Tasks

xAI released a new model, “Grok 4.5.” Purpose-built for coding, agentic tasks, and knowledge work, it was co-developed with Cursor and trained on large-scale data spanning coding, science, engineering, and mathematics.

Details

• Benchmarks: Achieved a 64.7% resolve rate on SWE Bench Pro, and 29.0% on SWE Marathon — the top score among competitors
• Token efficiency: Handles equivalent tasks using roughly 4.2x fewer tokens than Opus 4.8. Processing speed of 80 tokens per second puts it on par with fast models
• Strengths: Handles a wide range of tasks, from complex coding work involving Rust and C/C++ to office document creation in Excel, PowerPoint, and Word

Pricing (per million tokens)

Item	Price
Input	$2
Output	$6

How to try it

• Available immediately via Grok Build, Cursor (all plans), and the xAI API (console.x.ai)
• A free trial is also available for a limited time on Grok Build and Cursor
• Availability in the EU is expected to be delayed until mid-July