Introducing Grok 4.6 for reasoning and coding

Grok 4.6 is now available, bringing extended training runs and enhanced reasoning capabilities to software engineering and complex knowledge work.

Training and technical updates

Grok 4.6 underwent a longer supplemental training run than its predecessor, incorporating curated model-generated data for reasoning, advanced technical concepts, and high-quality engineering data. This foundation was followed by supervised fine-tuning, known as SFT — a training phase where the model learns from curated example outputs, and reinforcement learning stages using agentic tasks across domains like STEM and software development.

Developers used Grok 4.5 to regenerate SFT trajectories across multiple domains, filtering out problematic traces with model-based checks. The resulting checkpoint shows stronger performance in multi-step workflows, visual and interactive projects, and self-testing behaviors where the model checks its own work before proceeding.

Availability and pricing

The model is available today in Cursor and Grok Build, as well as through the API and partner platforms including OpenRouter, Vercel, and Cloudflare. Pricing starts at $2 per million input tokens and $6 per million output tokens, with a fast variant offered at twice that price. Double included usage is available inside Grok Build and Cursor for the first week.