Introducing Grok 4.5
Summary
Cursor and SpaceXAI have released Grok 4.5, their most intelligent mixture-of-experts model, designed for complex, long-running tasks across software engineering, data science, finance, and legal work. Trained jointly with SpaceXAI, Grok 4.5 utilized trillions of tokens from Cursor data, capturing user interactions with codebases and software tools, alongside a broad mix of high-quality STEM tasks and research papers. The model's proficiency stems from reinforcement learning applied to challenging problems in realistic environments, teaching it to investigate, use tools, recover from errors, and verify results. A distributed agent system was developed to construct these difficult training environments at scale. Grok 4.5 is available today across Cursor's desktop, web, iOS, CLI, and SDK platforms, with individual and team plans including significant usage, doubled for the first week. Pricing is \$2/M input and \$6/M output for the base model, and \$4/M input and \$18/M output for a fast variant. New cybersecurity safeguards have also been implemented.
Key takeaway
For AI Engineers integrating advanced agents into complex workflows, Grok 4.5 presents a broadly capable mixture-of-experts model extending beyond software engineering. You should evaluate its performance on long-running, multi-step tasks requiring creative tool use in domains like data science or finance. Consider its base and fast pricing tiers, starting at \$2/M input tokens. Utilize the doubled usage during the first week to assess its fit for your specific automation needs.
Key insights
Grok 4.5 is a multi-domain mixture-of-experts model trained via advanced RL for complex, tool-using tasks.
Principles
- Broad training data enhances multi-domain proficiency.
- RL on hard problems teaches robust agent behaviors.
- Distributed agents scale complex environment creation.
Method
The model was trained using reinforcement learning on difficult problems within realistic environments, leveraging a distributed agent system to construct and refine these complex scenarios at scale.
In practice
- Apply Grok 4.5 to multi-step software or data science tasks.
- Utilize its tool-use for finance or legal workflows.
- Compare base vs. fast variant for cost/performance needs.
Topics
- Grok 4.5
- Mixture-of-Experts
- Reinforcement Learning
- AI Agents
- Software Engineering
- Data Science
Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, AI Scientist
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Cursor Blog.