At the same time, SpaceX has launched a heavyweight research and development plan, launching a new basic large-scale model trained from scratch every month for the rest of this year, rewriting the global AI industry update cycle at a monthly iteration pace, and adding the code data brought by the $60 billion acquisition of Cursor, officially launching a positive technological competition with OpenAI, Google, and Anthropic.
1、 The landing code capability of Grok 4.5 core parameters is supplemented by the acquisition of Cursor
Grok 4.5 is built on a 1.5 trillion parameter V9 basic model, supplemented by Cursor programming data for training. Its code reasoning and generation capabilities have achieved a leapfrog improvement, and its internal evaluation scores are close to or even partially surpass Claude Opus.
According to the screenshot report of Musk X's original article on June 29th, the bottom base of Grok 4.5 is a 1.5 trillion parameter V9 model. The development team will fully integrate the AI programming tool Cursor's full interaction data during the later supplementary training stage, specifically to enhance code debugging, engineering scripts, and large-scale project logic disassembly capabilities.
At present, the model is not open to the public, and testing windows are only open to internal employees of SpaceX's aerospace research and development and Tesla's autonomous driving business lines; The team continues to promote RL reinforcement learning optimization, with daily iterations of the self-developed Grok Build standardized testing benchmark, and the model scores continue to rise.
Jiang Cheng, an independent analyst in the AI industry, said, "Top companies in the industry generally update their basic models on a quarterly basis, only making minor adjustments and incremental upgrades. Musk's monthly new product plan trained from scratch is essentially a radical military strategy under the dual pressure of computing power and capital." He mentioned that xAI has always had shortcomings in the code field. After completing the acquisition of Cursor in June, it directly filled the data barriers of developer scenarios, giving Grok a differentiated competitive track. Traditional general models can only complete simple script writing, while Grok 4.5 relies on a massive amount of real programming samples to support the complete industrial software development process, accurately meeting the needs of enterprise developers.
2、 60 billion US dollars acquisition of cursor, construction of code AI complete technology loop
On June 16th, SpaceX signed an acquisition agreement to acquire the parent company of cursor, Anysphere, in a $60 billion all stock transaction. The transaction is expected to be completed in the third quarter of 2026. This acquisition is not a financial investment, but a core strategic layout that supports the Grok series model code capabilities.
Sina Finance's special report on mergers and acquisitions on June 27th shows that this $60 billion acquisition is a full equity swap with high breakup protection clauses. Cursor's annual recurring revenue exceeds $4 billion, and over 70% of Fortune 500 companies are using this tool to develop programs.
Grok 4.5, released only 12 days after the acquisition, directly integrates with a massive number of developer interaction samples from Cursor, confirming the synchronous advancement of technology integration between the two parties. In the past, OpenAI and Anthropic only reached short-term computing power leasing cooperation with Cursor, while Musk directly acquired and mastered the underlying data supply, thoroughly mastering the upstream resources of programming AI.
Developer ecology observer Lin Zhou said, "The $60 billion acquisition of a single developer tool is an unprecedented investment in the AI industry. Musk's focus is not only on end products, but also on a massive dataset of real human programming thinking, which is a high-quality training material that cannot be artificially synthesized in the short term. Relying on SpaceX's Colossus million level GPU supercomputing cluster, Cursor will be deeply embedded into the Grok Build testing system to achieve an integrated closed-loop system of model training, code development, and benchmark evaluation, connecting the complete link from developer data usage to large model iteration.
3、 The new model iteration plan has been released, completely rewriting the competition cycle of the AI industry
SpaceX officially announced that it will release a new basic big model trained from scratch every month for the remaining months of this year, abandoning the industry's common incremental fine-tuning mode and compressing the AI iteration cycle from quarterly to monthly, reshaping the global big model competition landscape.
OpenAI、 The current mainstream rhythm for Google and Anthropic is to update the basic model once every three months, with only minor feature patches pushed in between; SpaceX has clearly defined the "completely zero training" standard this time, which means that every month's new products have new underlying weights, new training datasets, and new parameter architectures, and research and development investment and computing power consumption are exponentially increasing.
At the capital level, there is sufficient support. The original xAI completed a $20 billion Series E financing in January 2026, with strategic capital injections from technology giants such as Nvidia and Cisco; In February, SpaceX completed the full acquisition of xAI and officially renamed it SpaceXAI in May. The cash flow from the aerospace business continues to support the investment in AI's massive computing power.
Global technology industry researcher Shen Ming analyzed: "There are two major prerequisites for the implementation of the new model plan: one is the sufficient computing power reserve brought by the 20 billion financing, and the other is the acquisition of vertical scenario data by Cursor. This combination of punches will force other manufacturers to accelerate their research and development pace, and the AI arms race will officially enter a high-intensity stage. In the short term, the R&D cost threshold of the industry will be significantly raised, and the survival pressure of small and medium-sized model manufacturers will continue to increase. Industry resources will further concentrate on top capital giants. In the long run, high-speed iteration will continue to raise the upper limit of the ability of general large models and vertical code models, which is beneficial for the overall breakthrough of global AI technology.
