We recommend using Grok 4.6 for improved instruction following and long-horizon agentic work. Grok 4.5 remains available.

Grok 4.5 is a joint model from Cursor and SpaceXAI for long-running tasks across software engineering and knowledge work. It combines the Grok base with continued training on trillions of tokens of Cursor data and reinforcement learning on difficult, realistic problems.

Strengths

  • Handles long-running tasks that require using tools, checking results, recovering from mistakes, and adjusting its approach.
  • Applies broad knowledge to software engineering, data science, finance, research, legal work, and other computer-based tasks.
  • Solves multistep tasks in under half the steps of comparable frontier models.

Effort levels

Grok 4.5 supports three effort levels: high (the default), medium, and low. Higher effort gives the model more time to think through difficult tasks. Use high for the hardest work; drop to medium or low when you want faster, cheaper responses on simpler tasks.

On the Start plan (India only), Grok 4.5 is fixed at medium effort in non-fast mode. Effort switching and Fast mode are available on Pro and higher plans.

Tools

Grok 4.5 has access to all agent tools when used with Cursor, including:

Learn more about how tools work and tool calling fundamentals.

Pricing

Grok 4.5 is part of the Cursor Models pool on individual and team plans. This pool also includes Grok 4.6 and Composer 2.5.

Standard on-demand usage is priced at $2/M input tokens and $6/M output tokens. The Fast variant is priced at $4/M input tokens and $18/M output tokens. All prices are per million tokens.