xAI's new model is built to read contracts, not chat. That's a subcontract risk scan priced in cents, not a legal bill.

xAI released Grok 4.5 to the public on July 9, pitching it not as a chatbot but as a tool for "coding, agentic tasks, and knowledge work" — with legal contract review and financial document analysis named specifically as target use cases. It's priced at $2 per million input tokens and $6 per million output tokens, well under Anthropic's Claude Opus 4.8 ($5/$25) and in the same range as OpenAI's cheapest GPT-5.6 tier. That pricing puts AI-assisted contract review within reach of firms that could never justify licensing dedicated software for it.

What is Grok 4.5 actually for?

It's the first model xAI built jointly with Cursor, the coding tool SpaceX agreed to acquire for $60 billion in June. Grok 4.5 runs on xAI's new 1.5-trillion-parameter V9 foundation model, with training data pulled from Cursor's development workflows plus, per xAI, tasks spanning "software engineering, data science, finance, legal work, or anything else you do on a computer." In the legal category specifically, xAI's own materials list contract review, case file summarization, document comparison, and regulatory analysis as capabilities. That's a different pitch than a general chatbot — it's aimed at the same document-heavy grind that generates billable hours at a law firm or a risk-review desk at a GC.

Where does this land in construction?

Trimble already validated AI contract review as a real category when it acquired Document Crunch this spring — a tool that ingests a prime contract, subcontract, and exhibits and flags indemnity clauses, liquidated damages, no-damage-for-delay language, and notice-period deadlines clause by clause. That's a purpose-built, construction-trained product sold as enterprise software. Grok 4.5 is not that — it's a general-purpose model with no construction-specific tuning — but xAI is charging general-API rates for a capability the market has so far treated as premium software.

Run the token math and the gap is stark:

A 30- to 50-person mechanical or electrical sub that gets handed a new prime-to-sub agreement on every bid, and can't justify an enterprise contract-risk platform or a standing relationship with outside counsel for every job, is exactly the gap this pricing opens up. A first-pass scan for the clauses that actually shift risk — indemnification scope, pay-if-paid versus pay-when-paid language, consequential-damages waivers, insurance and COI requirements, notice deadlines for claims — costs closer to a rounding error than a legal invoice.

What this doesn't fix

xAI has not submitted Grok 4.5 to any independent benchmark — not LMArena, not Artificial Analysis, not any third-party evaluator. The only performance claims so far come from SpaceX and Tesla's own engineers, who are employees of xAI's parent company. That matters more here than it would for a coding task: a model that mis-reads an indemnification clause or misses a pay-if-paid trigger doesn't fail loudly like broken code does. It fails quietly, in a contract someone already signed. CAB flagged the same consolidation risk when SpaceX agreed to buy Cursor in June — construction teams building on that stack should already be tracking their exposure; Grok 4.5 is the first product of that deal reaching the public.

The takeaway

Treat a model like this as a flagging pass, not a legal opinion: run the subcontract through it, get a list of clauses worth a second look, then send that shortened list to whoever actually reviews contracts at your firm — in-house, outside counsel, or the owner reading it themselves. The cost of running it is now close to nothing. The cost of trusting it unchecked on a liability clause is not.

Forward this to the person on your team who's still arguing AI is overhyped.

Construction AI Brief publishes new analysis three times a week. Subscribe at constructionaibrief.com.