SpaceXAI launched Grok 4.5 on Wednesday, the company's first AI model built specifically for coding and autonomous agents, priced at $2 per million input tokens and $6 per million output tokens. The launch arrives one day before OpenAI plans to publicly release GPT-5.6, a convergence that puts two of the industry's most closely watched models into the market within 24 hours of each other.
Elon Musk framed the Grok 4.5 positioning on X.
Based on strong positive feedback from customers in our beta test program, @SpaceXAI will make Grok 4.5 available to the public tomorrow.
— Elon Musk (@elonmusk) July 8, 2026
It is an Opus-class model, but faster, more token-efficient and lower cost.
Anthropic's Claude Opus 4.8 is priced at $5 per million input tokens and $25 per million output tokens. OpenAI's GPT-5.6 Luna is priced at $1 per million input tokens and $6 per million output tokens.
The model is immediately available through Grok Build, the company's AI coding agent, through Cursor on all plans, and through the SpaceXAI developer console with an API key. EU availability is expected in mid-July.
What the benchmarks actually show
SpaceXAI trained Grok 4.5 across tens of thousands of Nvidia GB300 GPUs with a focus on data filtering, deduplication, and quality scoring. Benchmarking firm Artificial Analysis ranked the model fourth on its GDPval-AA v2 index with an Elo score of 1543, behind the latest Claude releases from Anthropic. The firm measured Grok 4.5 at $0.49 per completed task, "nearly 90% cheaper than the models ahead of it on our leaderboard," Artificial Analysis wrote, placing it "clearly on the Pareto frontier for performance versus cost."
On SpaceXAI's own published benchmarks, Grok 4.5 scored 62.0% on DeepSWE 1.0, behind Fable at 66.1% and GPT 5.5 at 64.31%, but ahead of Opus 4.8 at 55.75%. On SWE Marathon, Grok 4.5 led with a 29.0% resolution rate, above Opus 4.8 at 26.0% and Fable at 24.0%. Token efficiency tells a separate story. Grok 4.5 resolved SWE Bench Pro tasks with an average of 15,954 output tokens. Opus 4.8 used 67,020, roughly 4.2 times more tokens per task.
SpaceXAI serves the model at 80 tokens per second, which the company describes as fast-model speed for a frontier model.
Grok 4.5 is built for real-world engineering. It excels in large codebases and handles long-running tasks that span multiple repositories, hundreds of skills, and a variety of tools. pic.twitter.com/70qNssFqEz
— SpaceXAI (@SpaceXAI) July 8, 2026
The Cursor acquisition and what it contributed to training
Grok 4.5 is the first product to emerge from SpaceX's $60 billion acquisition of Anysphere, the company behind Cursor. SpaceX announced the all-stock deal days after its Nasdaq IPO in June.
"We've partnered with SpaceXAI to train Grok 4.5," Cursor said in a post on X on Wednesday.
We've partnered with SpaceXAI to train Grok 4.5.
— Cursor (@cursor_ai) July 8, 2026
It’s our most powerful model yet and the first we've built for more than software engineering. pic.twitter.com/U4B8Tedl34
The acquisition was structured in stages. In April, SpaceX took a right to buy Cursor for $60 billion, with a penalty of billions in fees and compute if it chose not to exercise that right, as Business Insider reported. Cursor's platform generates high-volume interaction data from engineers working on production codebases, and Musk said earlier this year that Cursor data had been incorporated directly into Grok's training.
The troubled path that preceded this launch
The polished launch comes after a difficult stretch for the organization behind the model. Grok generated antisemitic content in mid-2025 and at one point called itself "MechaHitler," incidents covered by HODLFM. Earlier this year, its image-generation feature allowed the creation of sexualized deepfakes that prompted investigations by the European Commission and the UK's Ofcom, as the BBC reported. SpaceX listed the behavior as a business risk in its own IPO filings.
All 11 of Musk's xAI co-founders had departed by the end of March, according to TechCrunch. Musk publicly acknowledged the organization had structural problems. "xAI was not built right the first time around," he said, stating he was rebuilding it from the foundations up. At a conference this spring, Musk said Grok was "currently behind in coding," a concession that set the specific terms for what Wednesday's launch needed to demonstrate.
SpaceX absorbed xAI in a share-exchange merger in February in a deal that CNBC confirmed valued the combined entity at $1.25 trillion, the largest merger on record. The June IPO followed as the largest in history. SpaceX shares rose 16% on the announcement of the Cursor deal.
Anastasiia G
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Please note that, despite the nature of much of the material created and hosted on this website, HODL FM operates as a media and informational platform, not a provider of financial advisory services. The opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice, HODL FM strongly recommends contacting a qualified industry professional.
