On July 9, Beijing time, SpaceXAI officially released a brand-new artificial intelligence model, Grok 4.5, developed in collaboration with the AI coding startup Cursor. This is the first AI model jointly trained by the two companies, marking a key step for Musk's AI business to expand from the programming field into the high-value legal and financial services markets.
Starting in programming, expanding to law and finance
Unlike Cursor's previous models that focused on programming, Grok 4.5 has a broader positioning, explicitly targeting legal services and financial services as application scenarios.
The model is designed for handling "complex and long-duration tasks," including software engineering, legal document analysis, financial data processing, and other high-difficulty work.
Cursor stated in its official blog that Grok 4.5 is "the most intelligent model to date, and the first built not only for software engineering," capable of "creatively using tools to solve problems, whether in software engineering, data science, finance, legal work, or other tasks you perform on your computer."
The model also adds new cybersecurity capabilities, along with corresponding protective measures.
Mixture of Experts architecture, trained on trillions of tokens
According to Cursor's official disclosure, Grok 4.5 is a Mixture of Experts model, jointly trained by Cursor and SpaceXAI.
The training used trillions of tokens of Cursor data, which capture extensive user interactions with codebases and software tools, allowing the model to learn both from existing software and from the interactions between developers and agents.
Unlike the earlier programming-specialized Composer 2.5, the development team deliberately made Grok 4.5's training data composition more extensive, incorporating high-quality STEM tasks, research papers, and other knowledge work content.
The model also underwent reinforcement learning in real-world environments covering software engineering and broader knowledge work, training it to explore problems, use tools, recover from errors, and verify results.
In terms of pricing, the base Grok 4.5 model costs $2 per million input tokens and $6 per million output tokens, with a fast variant also available. Both individual and team subscription plans include substantial usage of the model, with double usage in the first week.
Acquiring Cursor and making up for programming shortcomings
This release comes just weeks after SpaceX announced its acquisition of Cursor at a $60 billion valuation. Cursor is considered one of the fastest-growing AI startups in the "Vibe Coding" wave.
Musk publicly admitted earlier this year that his AI business had fallen behind competitors in code generation capabilities, and subsequently carried out large-scale personnel adjustments to rebuild relevant teams.
SpaceXAI only launched its first AI programming agent in May this year, later than competitors such as Anthropic.
This collaboration with Cursor is a key step for SpaceXAI to fill its shortcomings and attract enterprise customers.
Aiming at Wall Street, expanding into the enterprise market
The release of Grok 4.5 comes at a time when SpaceXAI is actively seeking to acquire customers on Wall Street, and the introduction of finance-specific capabilities is highly aligned with that strategic direction.
By sharing data and computing resources, SpaceXAI not only quickly made up for its previous shortcomings in programming but also explicitly extended its application scenarios to the high-value legal and financial services sectors.
At the same time, the release of the new model also coincides with a sensitive period in which the U.S. government is strengthening its review of next-generation AI models.
Cursor stated in its blog post that it has taken measures to "identify and block malicious actors," with the goal of "protecting legitimate security work, including discovering and patching vulnerabilities, while limiting the usage methods most likely to cause harm."
Currently, Grok 4.5 is officially available on Cursor's desktop, web, iOS, CLI, and SDK.