Grok 4.5 lands at $2/$6 with an ‘Opus-class’ claim and no receipts
Musk says Grok 4.5 is 'roughly comparable' to Anthropic's flagship at $2/$6. The evidence: company charts, an internal assessment and no system card.
SpaceXAI released Grok 4.5 on Wednesday, priced at $2 per million input tokens and $6 per million output, and available in Grok Build, the company’s API console, and every Cursor plan — everywhere, that is, except the EU, where the launch post promises availability by mid-July. Axios first reported the release Wednesday evening. Elon Musk’s pitch rests on a single comparison, and he made it twice.
“It is an Opus-class model, but faster, more token-efficient and lower cost.”
That’s Musk, per TechCrunch, invoking Anthropic’s flagship line, which charges $5 and $25 for its tokens. He went further: the company’s “internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster.”
An internal assessment is not a system card, and SpaceXAI did not publish one. What the launch post does contain is a deck of company-drawn benchmark charts on coding and agentic tasks, a claimed 80 tokens per second, and a token-efficiency comparison against Opus 4.8. What it doesn’t contain: a parameter count, evaluation methodology, or any third-party testing. The headline claim — Opus-class — currently cannot be checked by anyone outside the company. That’s the whole caveat; it applies to every capability statement in this story.
The timing is not subtle. OpenAI ships GPT-5.6 widely today, Thursday, a model TechCrunch describes as its strongest yet. Grok 4.5’s pricing brackets the field from below: TechCrunch notes OpenAI’s Sol runs $5 and $30, its cheaper Luna tier $1 and $6.
What’s under the hood, reportedly
The launch post says little about the model itself, but Musk has. In X posts on June 28, reported by GIGAZINE, he described Grok 4.5 as “based on our 1.5T V9 foundation model, with Cursor data added in supplemental training,” then in private beta at SpaceX and Tesla. Those figures come from Musk alone; the company hasn’t documented them anywhere formal.
The Cursor detail is the strategic one. SpaceX agreed in June to acquire Anysphere, Cursor’s parent, in a $60 billion all-stock deal — Quartz reported the merger filing, with closing expected in Q3 pending regulators. Quartz also noted Cursor had trained its newest model on tens of thousands of xAI chips, and that xAI itself was folded into SpaceX earlier this year. Buy the coding tool, feed its usage data to the model, ship the model back into the tool on day one. Grok 4.5 is the first release to complete that loop.
A lab with a ticker now
It’s also the first major model from a lab with a stock price to defend. That stock had a rough week already: SPCX shed 7.8% on July 2 after The Wall Street Journal reported the company had shown investors prototypes of a consumer AI device — a report Musk called “utterly false” on X, without elaborating.
Our read: that context explains the launch’s shape better than any benchmark does. A private xAI could let a model speak for itself over weeks of developer testing. A public SpaceXAI, days off a 7.8% drawdown and hours ahead of a rival’s flagship, has every incentive to announce first and document later. The costs of a thin launch land on whoever writes the next earnings call; the benefits land today.
The most precise number in Wednesday’s announcement is 15,954 — the output tokens the company says Grok 4.5 spent on tasks where Opus 4.8 spent 67,020. On the question of what the model actually is, precision is promised later.
Researched and drafted with AI tools, fact-checked against primary sources and approved by a human editor before publishing. Prices and dates are as of publication. Our standards →
Sources- x.ai — launch post
- axios.com — Axios first reported
- techcrunch.com — per TechCrunch
- gigazine.net — reported by GIGAZINE
- qz.com — Quartz reported
- tipranks.com — shed 7.8% on July 2
Latest from prompt/power
- Nvidia beats its own forecast with US$115.8 billion, then guides margins lowerNov 18
- Apple, the NSA and Chase: what each says broke in a week of outagesNov 18
- ChatGPT for Healthcare now has agents that click through hospital software. The fine print is all AmericanNov 17
- ChatGPT Wallet lets an AI agent pay for you. Here’s who can use it, and why Canada and the U.K. waitNov 17
- Gemini 4 Argon’s API price doubles Dec. 1. Here’s what it costs in US$, CA$ and £Nov 17