Grok 4.5 released for free

· Source: Machine Learning on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

SpaceXAI has released Grok 4.5, its newest flagship language model. It targets software engineering, coding agents, knowledge work, and enterprise productivity. The model was trained on tens of thousands of NVIDIA GB300 GPUs. It used large-scale Reinforcement Learning and hundreds of thousands of software engineering tasks. Grok 4.5 prioritizes real engineering productivity over raw benchmark numbers. It shows competitive performance across coding benchmarks. These include Terminal Bench 2.1, DeepSWE, and SWE Marathon, where it leads. On SWE-Bench Pro, it outperforms GPT-5.5. A key claim is its efficiency. It delivers around 80 tokens per second. Grok 4.5 achieves 4.2x fewer output tokens than Claude Opus 4.8 on SWE-Bench Pro. This leads to lower API costs and faster responses. Grok 4.5 also generates complex office documents. It is available via Grok Build, Cursor, and the SpaceXAI API. Limited-time free usage is also offered.

Key takeaway

For AI Engineers and ML teams evaluating coding models for agentic workflows, Grok 4.5 presents a compelling option. You should consider its strong software engineering capabilities, high inference speed (80 tokens/second), and significantly lower token consumption. Its aggressive pricing and optimized training for real-world programming are also key. Explore its free usage options in Grok Build or Cursor. Assess its fit for your specific multi-step coding and enterprise productivity tasks.

Key insights

Grok 4.5 prioritizes real-world engineering productivity and efficiency over pure benchmark dominance.

Principles

Method

Grok 4.5 was trained using large-scale Reinforcement Learning. It leveraged hundreds of thousands of software engineering tasks. Training included long-running asynchronous agent rollouts with high-quality filtered datasets.

In practice

Topics

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning on Medium.