Cursor is the software tool that most large companies' developers use to write and edit code with AI help. Think of it as a very smart assistant that sits inside the coding environment, reads the whole project, and can make changes across dozens of files at once, following plain instructions. It is used by companies ranging from Nvidia, where all 40,000 engineers reportedly use it, to Coinbase. On May 18, Cursor released Composer 2.5, its own AI model that sits at the heart of this tool. The key practical improvement is that it handles long, multi-step jobs better. The previous version would sometimes lose track of instructions or do too much or too little work on a task. Version 2.5 was trained to calibrate its effort to the size of the problem and follow complex instructions more consistently. The pricing is the other headline. The standard tier costs $0.50 per million words of input, and $2.50 per million words of output. For comparison, Anthropic's Claude Opus 4.7 is priced at roughly $15 input and $75 output per million tokens. Benchmark scores put Composer 2.5 in a similar performance range to those frontier models at about one tenth the cost. Whether that holds in real-world use on messy, legacy codebases remains to be tested independently. For business leaders, the Cursor story carries a thread they need to know about. The model is built on top of Kimi K2.5, a Chinese open-source AI model from Moonshot AI, backed by Alibaba. When Cursor launched its previous model Composer 2 in March 2026, it did not disclose this. A developer found the Chinese model's name hidden in the system's own API responses. Moonshot AI publicly accused Cursor of violating its license terms, congressional committees sent formal inquiry letters to Cursor over its Chinese AI dependency, and the incident triggered a broader industry conversation about whether companies know what is actually processing their data. Cursor has been upfront about the Kimi foundation in this release, and the authorized commercial arrangement with Moonshot AI is confirmed. But the model has the same Chinese lineage as before. For companies operating under data sovereignty requirements, GDPR, HIPAA, government contracts, or any rules about where sensitive information can be processed, this is worth checking directly with your IT and legal teams. The cleaner solution is coming. Cursor announced a partnership with SpaceXAI to train an entirely new model from scratch, using ten times more computing power than Composer 2.5. That model would be built without a Chinese foundation. SpaceX has also announced an option to acquire Cursor outright for $60 billion before the end of 2026. SpaceX can instead pay $10 billion for the collaboration work if it decides not to buy. Either outcome would significantly change Cursor's ownership and strategic direction. For businesses that already use Cursor, nothing changes operationally today. Composer 2.5 is a better version of what was already there. For businesses evaluating whether to adopt it, the pricing advantage is real and the transparency question has been partially addressed. The bigger decision is whether to treat the current Chinese model base as an acceptable short-term situation while the next model is in training, or to hold back and wait.
Product Launch2 min read
Cursor Launches Composer 2.5, Its Own AI Coding Model
June 5, 2026Synthesized from 1 source: TLDR AI
Cursor, the AI coding tool used by over half of Fortune 500 companies, has released Composer 2.5, a significantly improved version of its own AI model that handles long, complex coding jobs at a fraction of what rival models charge, while also disclosing a potential $60 billion acquisition by SpaceX and carrying unresolved questions about its Chinese AI foundation.
Related Coverage
Stay informed
Get AI intelligence like this delivered to your inbox.