SpaceX’s Grok 4.5 launches at half the price of rivals – which is why it could oust Anthropic and OpenAI
Elon Musk’s SpaceX released the Grok 4.5 on Wednesday, the first intelligence model the company has specially trained to code and autonomous agents — and the first visible product of its $60 billion AI startup coding, completed a few weeks ago.
The launch marks an important test of the widespread AI empire, which Musk has been putting together for the past six months, and the strategy that betting developers care less about topping leaderboards than about speed, cost, and whether the model can do the job.
"Announcing Grok 4.5, our first model specifically designed for coding and agents," the company said in a post to X. "It was trained with Cursor and provides borderline intelligence with leading speed and cost efficiency."
Why the Grok 4.5’s pricing strategy is more important than its benchmark score
SpaceX does not claim that the Grok 4.5 is the smartest model in the world. Instead, make an economic argument. The company claims that the model uses half as many tokens per transaction as comparable models, delivers high output, and costs less than most – it costs $2 per million input tokens and $6 per million output tokens. That undercuts high-end competitors like Anthropic’s Claude Opus line and OpenAI’s borderline models by a wide margin.
Musk has laid out the framework for a sincere stance. "Our internal evaluation is that the Grok 4.5 is almost comparable to the Opus 4.7, but much faster," wrote to X. "The combination of capability, fast speed and low cost is what makes it competitive. We close the loop on real-world usability, not benchmarks. Hardcore engineers at Tesla & SpaceX find Grok 4.5 really useful, which is what really matters."
That framework is both a philosophy and a hedge. An independent test released on Wednesday suggests the Grok 4.5 has real competition but isn’t overpowering. Ratings firm Artificial Analysis ranked the model fourth in its GDPval-AA v2 index of real-world agent knowledge performance, with an Elo score of 1543, "after Claude’s latest release from Anthropic." But the cost figures are where the model shines. Transaction Analysis rated Grok 4.5 at $0.49 per completed transaction — "almost 90% cheaper than its predecessors on our leaderboard," the company wrote, and put it "clearly on the Pareto frontier in performance versus cost."
For business buyers, those statistics are very important. Agenttic workloads – where the model runs automatically for minutes or hours, learning code bases, calling tools, and iterating on its output – consumes a lot of tokens. A model that is 90% cheaper for completed work, even if it is less skilled, changes the calculation of any engineering organization that sends agents to hundreds of engineers. Investor Gavin Baker holds a cautious outlook for the market: "Pareto best in numerical coding. We will see in the most important vibes."
How 60 billion Cursor acquisitions shaped Grok 4.5 training
Grok 4.5 is the first physical evidence of what SpaceX bought when it acquired Cursor, and the deal itself happened in phases. In April, SpaceX made an unusual offer that gave it the right to buy a coding startup for $60 billion — or pay billions in cash and count when we leave, as Business Insider reported at the time. Days after SpaceX’s Nasdaq listing in June, the company exercised that right, announcing a stock acquisition that CNBC reported was a 3.4% dilution of the IPO estimate. SpaceX shares rose 16% on the news.
The logic of the strategy was always about data as a product. Cursor’s AI-first code editor generates massive streams of high-quality interactive data: how professional engineers write, edit, review, and modify code in real production environments. Musk made it clear this spring that Cursor’s interactive data was being fed directly into Grok’s training. Cursor, in turn, gained access to SpaceX’s Colossus supercomputer in Memphis – about 200,000 Nvidia GPUs with plans to reach one million – after publicly admitting that "blocked by computing."
"We partnered with SpaceXAI to train Grok 4.5," Cursor’s official account was posted on Wednesday. "It’s our most powerful model yet and the first one we’ve built beyond software engineering." SpaceX says the model shows that pedigree: it "thrives on large codes and handles long-running tasks involving multiple repositories, large capabilities, and various tools" — precisely the messy, multi-file reality of professional software engineering that clean code benchmarks often fail to capture. Early reactions from engineers suggest that the training has paid off. "Ok Grok 4.5 is a waste," posted by developer Evan Bacon. "Just made this rocket tracking app with live data and 3D globe. I might need a new benchmark after this."
Inside xAI’s tumultuous year of scandals, departures, and restructuring
The polished presentation belies how chaotic the road was. Grok has spent much of the past year struggling. In mid-2025, the chatbot produced antisemitic content and at one point called itself "Mecha Hitler," episodes covered extensively by NPR and CNN. Earlier this year, its image generation features allowed users to deepfake sex, including children – drawing investigations from the European Commission and Britain’s Ofcom, as the BBC reported, and prompting SpaceX to list the behavior as a business risk in its IPO filing.
The organization behind the model also broke. All 11 of Musk’s xAI co-founders left at the end of March, according to TechCrunch, and Musk publicly acknowledged that xAI. "it is not well constructed [the] first round," saying he is rebuilding them "from the foundations up." Musk himself admitted at a conference this spring that Grok was there "currently behind in coding" – rare public approval from anonymous administrators.
Against that background, Grok 4.5 reads like the first product of a reinvented organization — and the first point of proof of the bold story SpaceX told public market investors. At the time of its IPO, the company estimated a market cap of $28 trillion, with nearly $26 billion tied to AI, including $22.7 trillion. "business applications" opportunity. Those numbers are hard to believe even by Silicon Valley standards. A competitive, low-cost coding model is the most direct route from that narrative to real revenue, which is why Wednesday’s launch carries more weight than traditional model releases.
Grok 4.5 vs. Claude: AI code market war
The competitive stakes are hard to overestimate, because the AI coding market has always featured one leader — and it’s not Musk. Even as Cursor’s revenue exploded, its market share was eroding. A breakdown of data from Ramp cited by CNBC showed Cursor’s share in the AI coding segment falling from 41% in June 2025 to about 26% in May 2026, while Anthropic controls nearly half of the market. Anthropic also topped CNBC’s Disruptor 50 list this year and, according to Artificial Intelligence, still holds the top spot in agent performance ratings.
That’s the gap the Grok 4.5 is designed to close – not by thinking outside of Claude, but by reducing it. The economics of this model create classic disruptive disruption: if it brings more power to the border at a fraction of the cost per transaction, price-sensitive business loads will migrate, and incumbents will face pressure on their most profitable API traffic. The counterargument is that in coding, qualitative combinations. A model that solves a complex bug correctly on the first try would be cheaper to run than one that costs half as much per token but requires three tries. Hence Baker’s caveat about "vibes" – the engineering community’s summary of a sound model of reliability in real work – will determine more than any benchmark the launch date.
There is also a structural question buried in the agreement. Cursor built its business on offering developers their choice of models, including Claude and GPT. If Grok becomes the favorite child inside Cursor – and Musk was already urging users to do so "Try Grok 4.5 on Cursor!" within hours of launch – the product risks alienating the very users whose data made Grok 4.5 possible. Regulators, already scrutinizing Grok for security reasons in two areas, may be more interested in a company that manages training data, a model, and a high-level distribution channel at the same time.
What Musk’s trillion-dollar direct integration bet means for the future of AI
Grok 4.5 also clarifies what Musk’s chaotic engagement is building towards. In February, SpaceX absorbed xAI in a transaction that CNBC confirmed valued the combined company at $1.25 billion — the largest merger ever, valuing SpaceX at $1 trillion and xAI at $250 billion. A June IPO followed, the largest in history, and the stock soared past $200 from its $135 offering, knocking SpaceX past Amazon and Microsoft to become the fourth most valuable company in the United States.
The result is a single public company that owns almost the entire stack: Colossus to hold training, the desires of orbital data centers to enable future scaling, the frontier model at Grok, the distribution channel at the base of the Cursor engineer, and the need for capture from the engineering organizations of Tesla and SpaceX. Neither OpenAI nor Anthropic can replicate that integration; both must reach developers through third-party tools, some of which Musk now owns. Whether that concentration proves to be an unassailable drain or target — or both — is now one of the defining questions in the AI business.
In the next few weeks you will begin to answer it. Artificial Analysis says its full Intelligence Index results are forthcoming. Business pilots will reveal whether the claims of token usage survive the connection to the actual codes. And Anthropic, which has responded to every major challenge of this cycle with rapid counter-releases, is unlikely to withdraw the price action margin quietly.
But the deeper story of Grok 4.5 may be what it says about where the AI race has moved. For three years, the scoreboard in the industry was intelligence: whose model was the smartest. Musk, latecomer and beaten, chose to compete on a completely different axis – its very cheap model to use. It’s a good decision from a man who made his fortune not by inventing the rocket or the electric car, but by relentlessly driving up the cost of doing it. If the strategy works, Musk will have done for AI what he did for space flight. If it doesn’t, he will have spent $60 billion to learn that in software, unlike rockets, the cheapest ride is not always the choice of engineers.
