xAI launched Grok 4.5 on July 9, 2026, the first major model release since SpaceX completed its acquisition of xAI earlier this year. The model targets coding, research, and routine knowledge work. It is live now in Cursor, Grok Build, and the xAI API console. EU availability is coming mid-July.
The pricing is the most concrete fact here. At $2 per million input tokens and $6 per million output tokens, Grok 4.5 costs less than half of what Anthropic charges for Opus 4.8 ($5 input, $25 output), and a fraction of what Fable 5 runs ($10 input, $50 output). For a business running AI across hundreds of documents, emails, or reports per day, that gap compounds fast.
xAI also claims the model uses 4.2 times fewer tokens than Opus 4.8 to complete the same software engineering tasks. Tokens are the units of text an AI model reads and writes; you pay per token, so using fewer of them to do the same job lowers the bill further. If that figure holds in practice, Grok 4.5 would be dramatically cheaper to run than even the sticker price comparison suggests.
Benchmarks paint a mixed but honest picture. On Terminal Bench 2.1, which tests complex command-line tasks, Grok 4.5 scores 83.3%, essentially matching GPT-5.5 at 83.4% and just behind Fable 5 at 84.3%. On deeper software engineering tests, the gaps widen: Grok 4.5 scores 53% on DeepSWE 1.1, versus 67% for GPT-5.5 and 70% for Fable 5. These numbers came from xAI's own launch materials. Independent third-party verification does not exist yet, which is normal for a same-day release but worth keeping in mind.
One result that stands out: Grok 4.5 reportedly scores first on Harvey's Legal Agent Benchmark, a test built around legal reasoning tasks. A model positioned primarily around coding topping a legal benchmark suggests the training mix reaches wider than xAI's marketing implies. That matters for professionals in law, compliance, procurement, and finance who might otherwise assume this release is only relevant to software teams.
The model was trained alongside Cursor, an AI coding tool that SpaceX agreed to acquire for roughly $60 billion in June 2026. That deal is expected to close in Q3. The practical effect is that Grok 4.5 was trained on real developer session data from millions of Cursor users: actual debugging traces, multi-file edits, and corrections. That kind of grounded training data is harder to replicate from synthetic sources, and it may explain why the model performs well on practical coding tasks even where it trails on abstract benchmarks.
The pricing strategy mirrors what Chinese AI companies like DeepSeek have done: get close enough on performance, then compete heavily on cost. xAI is doing the same within the Western AI market. For business operators who are already paying for AI tools through subscriptions or API access, Grok 4.5 is worth evaluating as an alternative, especially for high-volume, repetitive tasks like document summarisation, data extraction, or research drafts where raw quality differences between top models are less noticeable than cost differences.
The model is not yet available in the EU, which limits its relevance for European operators right now. That window is expected to open by mid-July. For everyone else, xAI is offering free usage in Grok Build and Cursor for a limited time, which is a low-cost way to test it against whatever you are currently using.