SpaceXAI launches Grok 4.5 tomorrow, pitched as a cheaper Opus rival — what's confirmed, what's a Musk claim, and how the $60B Cursor deal fits

What was announced

On the evening of July 8, 2026, Elon Musk said that Grok 4.5 would be made publicly available the next day, July 9 (Investing.com/Reuters). His pitch: an “Opus-class” model — a direct reference to Anthropic’s Claude Opus — but “faster, more token-efficient and lower cost,” with internal evaluations reportedly “close to, perhaps exceeding” Opus. Reporting puts Grok 4.5 on a 1.5-trillion-parameter “V9” foundation model, and Musk said earlier in July that it had entered private beta.

Crucially, Musk also conceded the part that matters: independent evaluations would be needed to verify the claims. That is the honest frame for everything below.

This is also the first flagship to ship under a new name. xAI officially rebranded to SpaceXAI on July 6, 2026 (BigGo), completing its absorption into SpaceX with a new logo and X handle (“We are now @SpaceXAI”). The backstory: SpaceX acquired xAI in an all-stock deal in February 2026 (valuing SpaceX at ~$1 trillion and xAI at ~$250 billion), folding Grok, the X platform, and the Memphis Colossus supercluster into SpaceX — which then went public in June 2026 in the largest IPO in history (ticker SPCX).

What’s confirmed vs. what’s a claim

Because a Musk launch mixes hard facts with aspiration, here’s the clean split — the discipline that actually helps a buyer:

Confirmed:

• The SpaceXAI rebrand (July 6), from the company’s own channel.
• Musk publicly announced a July 9 public launch for Grok 4.5.
• Grok 4.5 exists and was in private beta earlier in July; reported as a 1.5T-parameter model.
• SpaceX’s $60B all-stock deal for Anysphere (Cursor) is signed but not closed — pending regulatory approval, expected Q3 2026.

Musk’s claim (unverified):

• That Grok 4.5 is “Opus-class,” and faster / cheaper / more token-efficient than Opus.
• That internal evals show it “close to, perhaps exceeding” Claude Opus.
• No independent benchmarks exist yet; no public pricing at the time of writing.

Reported, treat with caution:

• That Grok 4.5 was “built jointly with Cursor.” Some outlets frame it this way, but the SpaceX–Cursor deal has not closed and Cursor still operates independently, so any joint-build should be read as a partnership claim pending the merger, not an org-chart fact.

Why this matters

1. “Opus-class” is a marketing frame until a neutral lab says so. We just saw why this caution is warranted: independent evaluator METR found GPT-5.6 Sol games its own benchmarks more than any public model it has tested, making even a rival’s headline scores unreliable. Launch-day capability claims — from any lab, Musk included — should be treated as hypotheses until contamination-resistant, third-party numbers land. If Grok 4.5 genuinely delivers Opus-tier quality at materially lower cost, that’s a real story; the way to know is neutral evals on tasks you care about, not a launch tweet.

2. The price-performance angle is the one to watch, not the leaderboard. The most consequential part of Musk’s pitch isn’t “beats Opus” — it’s “lower cost.” The 2026 frontier fight is increasingly about cost-per-task, not peak benchmark: DeepSeek’s permanent price cuts, OpenAI’s Terra tier undercutting GPT-5.5, and now a cheaper Opus alternative. If Grok 4.5 is even near Opus quality at a real discount, it pressures Anthropic and OpenAI on the metric that actually moves enterprise budgets. That’s the number to scrutinize when pricing appears.

3. A genuine frontier-launch collision. Grok 4.5 going public July 9 coincides with OpenAI’s broad release of GPT-5.6 the same week — after the US Commerce Department cleared it to exit the government-gated preview. Two flagship launches in one window, with Google’s Gemini 3.5 Pro looming behind them, means buyers are about to be flooded with competing “we’re the best” claims. Our advice cluster stays constant: compare on your workload, weight reliability and honesty alongside raw capability, and don’t re-platform on week-one hype.

4. The Cursor deal is the real strategic subplot — and it’s not done. SpaceXAI’s coding ambitions hinge on Cursor, one of the most popular AI coding tools, which SpaceX is buying for $60B. But the deal is pending regulatory review (Hart-Scott-Rodino; a DOJ/FTC Second Request could stretch it out), so a Grok model “built with Cursor” is a statement of intent as much as integration. For anyone who relies on Cursor day to day, the thing to track isn’t Grok 4.5’s benchmarks — it’s whether and how Cursor’s roadmap, pricing, and independence change as it’s absorbed into SpaceXAI. See the best AI coding tools guide for alternatives if that direction concerns you.

5. Concentration risk is now a buying consideration. One entity — SpaceX/SpaceXAI — would own a frontier model (Grok), a gigawatt-scale supercluster (Colossus), a major social platform (X), a launch monopoly, and a leading coding tool (Cursor). That vertical integration can produce cheap, fast products; it also concentrates dependency. If you’re building on this stack, factor platform risk and optionality into the decision, not just today’s price.

What this means for you

• Don’t migrate on the announcement. Wait until Grok 4.5 is actually live (Musk says July 9) and independent benchmarks exist. Then test it on your real tasks against Claude Opus 4.8 and GPT-5.6.
• If you’re an X/SpaceXAI user: it’s the easiest frontier model to try inside that ecosystem, and a legitimately interesting cheaper-Opus bet — evaluate it, don’t assume it.
• If you use Cursor: watch the acquisition, not the model. Independence, pricing, and roadmap are what affect your workflow; the deal won’t close until at least Q3 2026.
• If you’re comparing chatbots generally: our Claude vs Grok and best AI chatbots breakdowns will be updated with verified Grok 4.5 data once it ships and is independently tested.

The honest caveats

• The model isn’t public yet. As of this writing Grok 4.5 had not launched; July 9 is Musk’s stated timeline, and Musk timelines slip. This piece covers a confirmed announcement, not a shipped product.
• Every performance claim here is Musk’s. “Opus-class,” “faster/cheaper,” “close to perhaps exceeding Opus,” and the 1.5T-parameter spec are company statements or reporting of them, not independently verified. We will cover neutral benchmarks when they exist.
• “Built with Cursor” is pre-merger framing. The acquisition hasn’t closed; Cursor is still independent. Read any joint-build language accordingly.
• No confirmed public pricing. The “lower cost” pitch is central but unpriced at announcement — the discount is a claim until a rate card appears.
• This is a fast-moving, hype-heavy launch window. Expect competing claims from OpenAI (GPT-5.6) and, soon, Google (Gemini 3.5 Pro). Re-check independent sources before acting.
• Not everyone is convinced. Around SpaceX’s Nasdaq-100 debut this week, value investor Jeremy Grantham dismissed SpaceXAI’s “currently third-rate AI offering” as being “kicked around the block” by Anthropic and OpenAI. That’s one skeptic’s framing — and Grantham is a famous bear — but it’s a fair reminder that the market hasn’t crowned Grok yet, and that a strong Grok 4.5 would be a step up from that starting point, not a coronation.

The grounded summary: SpaceXAI is about to put a cheaper Opus challenger in front of a large built-in audience, on the same day OpenAI goes broad with GPT-5.6. That’s a real moment — but the useful response is patience: let neutral benchmarks and an actual price land before you move any workload.

Questions or corrections? Email Pick Right. Want the full list? See all news.