xAI Launches Grok 4.5 With Pricing Built to Undercut Anthropic's Opus 4.8

XAI launched Grok 4.5 on July 9, pricing it at $2 per million input tokens and $6 per million output tokens to undercut Anthropic's Claude Opus 4.8. The model, built on a 1.5 trillion parameter mixture-of-experts foundation model and trained with data from coding assistant Cursor, targets developers and enterprise users in legal and financial work. Independent benchmarks have not yet confirmed Musk's claim that it matches Opus on quality, but the pricing war and distribution through Cursor's developer base are seen as strategic moves.

Elon Musk says Grok 4.5 matches Anthropic's top model for a fraction of the price, and the timing is no accident. Musk posted on X on July 8 that xAI would open Grok 4.5 to the public the next day, calling it "an Opus-class model, but faster, more token-efficient and lower cost." That's not a vague boast. It's a price quote aimed squarely at Anthropic. Grok 4.5 runs at $2 per million input tokens and $6 per million output tokens, against $5 and $25 for Claude Opus 4.8. A faster variant costs more, $4 and $18, but still lands well under Anthropic's number. For a developer burning through millions of tokens a day on an agentic coding tool, that gap is real money, not a rounding error. The model itself is new under the hood. Grok 4.5 sits on a 1.5 trillion parameter mixture-of-experts foundation model xAI calls V9, and it's the first Grok trained with data from Cursor, the coding assistant that SpaceX agreed to buy for $60 billion in an all-stock deal announced June 16, according to TechCrunch and Axios. That acquisition is expected to close in the third quarter. xAI, which merged into SpaceX earlier this year, built Grok 4.5 jointly with Cursor's team, and the training set reportedly includes trillions of tokens drawn from how real users work inside codebases, not just scraped GitHub repositories. The result is live right now in Grok Build, across every Cursor plan, and in the xAI console, ahead of what xAI is billing as a wider Thursday rollout. xAI is pitching the model beyond coding too, into legal and financial work, the kind of document-heavy, rule-bound tasks where a cheaper model that reasons carefully can replace a lot of billable hours. That's a real market. It's also exactly where Anthropic has spent two years building enterprise relationships, so the overlap is deliberate. Here's the thing worth saying plainly: independent testing doesn't yet support "beats Opus" as a blanket claim. Superconductor, which runs its own coding benchmark against a production Rails codebase, found Grok Build fast but not the strongest performer, writing that it is "not beating GPT-5.5 or Opus 4.7 on quality" while still being quick enough that the firm now includes it when generating multiple candidate fixes for hard tickets. That's a genuinely useful role for an AI model, the fast second opinion, but it's a step below the frontier-beating language Musk used on X. Grok 4.20, the prior flagship, scored 78% on SWE-bench Verified, trailing GPT-5.4 at 81.5% and Claude Opus 4.6 at 76%, depending on which leaderboard you check. Grok 4.5's own verified numbers weren't published alongside the launch post, which makes the Opus comparison hard to check independently right now. Frankly, the pricing war matters more than the benchmark war at this stage. OpenAI just put out its own answer. GPT-5.6 launched in three tiers in late June: Sol at $5 input and $30 output for the heaviest reasoning work, Terra at $2.50 and $15 as a mid-tier option, and Luna at $1 and $6 for high-volume, lower-stakes tasks. Comparing Grok 4.5 to Luna, as some coverage has done, isn't quite fair since Luna isn't OpenAI's frontier tier. The honest comparison is Grok 4.5 against Sol and Opus 4.8, and on that basis xAI is still the cheapest of the three labs claiming top-tier reasoning. What's changed is the distribution channel. Cursor now has hundreds of thousands of paying developers who touch a model selector every day, and xAI just bought its way into a default option there through the SpaceX deal rather than through a standalone marketing push. That's arguably a bigger deal for xAI's business than the model itself. Anthropic and OpenAI still sell mostly through their own APIs and consumer apps. xAI just acquired a distribution channel that sits inside the workflow of the exact customers it's chasing. Whether Grok 4.5 actually holds up against Opus 4.8 on real engineering work will get answered fast, since Cursor's own users will be running head-to-head comparisons within days of Thursday's wider release. The price gap won't close on its own. What could close it is Anthropic, still the incumbent in enterprise coding tools, deciding it has room to cut. Also read: Prime Intellect Raises $130 Million for Enterprise AI Agents https://startupfortune.com/prime-intellect-raises-130-million-for-enterprise-ai-agents/ • China Will Let Its Top AI Firms Buy Nvidia's H200 Chips Again https://startupfortune.com/china-will-let-its-top-ai-firms-buy-nvidias-h200-chips-again/ • Metal Turns Fundraising From a Networking Problem Into a Data Problem https://startupfortune.com/metal-turns-fundraising-from-a-networking-problem-into-a-data-problem/