OpenAI Terminates Cursor Partnership

OpenAI has announced the termination of its partnership with Cursor, stating that direct access to its models for Cursor users will end on November 12th. This decision follows Cursor's recent acquisition by SpaceX.

The announcement was made by Theo, who stated, "We're ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor's direct access to our models would end on November 12th."

This move impacts Cursor, a code editor that integrates AI, which will no longer be able to provide direct access to OpenAI's powerful language models.

The termination raises questions about the future of AI development tools and the autonomy of smaller companies in a rapidly consolidating market.

SpaceX acquired Cursor in a deal that was initially considered a joint partnership, with an option for a $60 billion buyout at year-end. However, SpaceX opted to expedite the acquisition, according to Theo.

The motivation behind the acquisition was to gain access to Cursor's valuable data and its engineers, particularly those specializing in Reinforcement Learning (RL). This strategic move is intended to directly improve the performance of Grok, SpaceX's proprietary AI model.

Theo confirmed his personal financial interest in Cursor, stating, "I was an early investor in Cursor and I've been getting spammed with emails and still have no idea if I'll even be allowed to get the stock, much less sell it."

He expressed uncertainty regarding the status of his equity and the process for selling his shares following the acquisition.

Theo clarified that he has no financial ties to OpenAI, stating, "I also have no tie financially whatsoever to OpenAI. In fact, I spend a lot of money on OpenAI, so take that as you will."

He further noted that he also spends money on Anthropic and tends to be more critical of companies with which he has financial connections, implying his impartiality in this matter.

OpenAI formally notified SpaceX of its intent to terminate the contract providing OpenAI models to Cursor. The proposed shut-off date for this access is November 12, 2026.

Theo quoted the notification: "Today we notified SpaceX that we intend to wind down our contract providing OpenAI models to Cursor with a proposed shut off date of November 12th, 2026." This indicates OpenAI's intention to strictly adhere to contract specifications in winding down the partnership.

Theo highlighted the high-profit margins associated with hosting AI models, stating, "the margins for hosting the models are really good for those businesses, which gives them a lot of wiggle room in how they negotiate these contracts."

This strong financial position provides companies like Google and OpenAI significant leverage in contract negotiations. Cursor was noted as a major consumer of Anthropic's inference services, at one point accounting for over 70% of Anthropic's non-research inference.

Anthropic's compute chief publicly expressed support for Cursor, signaling a strategic alliance between Anthropic and XAI against OpenAI. The statement read, "We'll continue to increase compute to support cloud models in Cursor and are excited for what comes next with them at SpaceX."

This collaboration is significant as Anthropic relies on SpaceX for some of its compute resources, creating a mutually beneficial relationship. The move suggests a growing alignment of interests among AI companies that may view OpenAI as a dominant competitor.

This alliance could reshape the competitive landscape in the AI sector, fostering new partnerships and potentially intensifying rivalries.

By cutting off Cursor, OpenAI is likely losing a significant revenue driver, as Cursor was a substantial user of their models. Theo noted, "Anyways, this also means OpenAI is probably losing a pretty big revenue driver by blocking Cursor."

It is speculated that Cursor paid a heavily discounted rate for OpenAI's services, unlike typical subscription models which are often subsidized. This implies that the revenue from Cursor was likely a direct and substantial contribution to OpenAI's finances.

OpenAI officially cited concerns over contractual violations and data usage as the reason for terminating the partnership. Theo quoted OpenAI's statement: "We are making this choice because we cannot be confident that SpaceX will use our technology within our terms of service based on our experience with Elon Musk's company's violating contracts."

OpenAI specifically fears that Elon Musk and SpaceX might be reverse-engineering or "distilling" their models. A primary concern revolves around the collection of input/output data from OpenAI's models being used to train Grok, which would provide a significant data advantage to SpaceX.

This concern is rooted in a history of perceived contractual breaches by Elon Musk's companies.

The termination underscores the fierce competition for data and model intellectual property in the AI industry, as companies strive to protect their proprietary technologies.

There are concerns that Cursor and xAI are collecting data from user interactions to train their own models. Theo stated, "There's a lot of customers who don't hit that, which ends up being massively useful for Cursor and now SpaceXi in collecting this data to train against."

This data collection is reportedly beneficial for providers, especially when customers opt into data sharing, allowing for continuous model improvement.

OpenAI previously relied heavily on Cursor, even providing early access to GPT-5 and detailed instructions for its setup within Cursor, recognizing it as the optimal testing environment. Theo recounted, "When they gave us the early access, they gave us a Google doc with detailed instructions on how to set up the model inside of Cursor because that was the best place to test it."

OpenAI engineers reportedly favored Cursor over their native Codex CLI for model testing and development. This history suggests a deeper, albeit now fractured, collaborative past between the two entities.

This previous reliance highlights Cursor's significant role in the development and refinement of OpenAI's models, making the current termination a notable shift in their relationship.

OpenAI has confirmed that the changes impacting Cursor will not affect T3's specific tools and projects. Theo stated, "I had a contact at OpenAI hit us up making sure we knew that this will not affect T3 code and that they plan to continue supporting projects like ours indefinitely going forward."

This reassurance indicates OpenAI's intention to maintain support for other third-party projects and integrations, despite the termination of the Cursor partnership.

Theo criticized Cursor's public response to the termination, labeling it as "unnecessarily dunky." He argued that Cursor's framing of the announcement as a total block on OpenAI models is misleading.

He clarified that users can still access OpenAI models within Cursor by utilizing them through the API or other alternative methods, rather than direct integration. Theo noted, "This is egregious wording in my opinion because you can still use OpenAI models in cursor if you use them through the API or other ways."

This suggests that Cursor's communication may have overstated the impact of the termination, potentially creating undue concern among its user base.

Token utilization is often used as a primary metric for AI model usage, but Theo argues it is a deeply flawed method for comparing different AI models, stating, "The biggest by far is token utilization."

OpenAI models are significantly more token-efficient than many competitors, such as Anthropic's Claude 3.5 Opus or Sonnet. Measuring traffic solely by raw token volume can inflate the usage figures for less efficient models, making them appear to have higher traffic than they actually do in terms of computational load.

This discrepancy can lead to misinterpretations of actual model usage and efficiency across different platforms.

Theo highlighted a substantial efficiency gap in agentic tasks, comparing OpenAI's o1 (Soul) with Anthropic's Claude 3.5 Sonnet (Fable). On the 'Artificial Analysis' benchmark, OpenAI's o1 uses approximately 17,000 tokens, while Claude 3.5 Sonnet uses about 36,000 tokens.

Cursor's own benchmarks reveal an even larger disparity, showing 28,000 tokens for o1 versus 103,000 for Claude 3.5 Sonnet. This significant efficiency gap means that a reported 5% of traffic attributed to certain models could, in reality, represent a 15% actual task load due to token inefficiency.

Theo critiqued OpenAI's '5%' traffic claim, stating that it needs a caveat to clarify whether it refers to requests, users, or tokens. He explained, "If anthropic models are three times less token efficient, then the 5% should be thought of more like 15%."

Because OpenAI's models are more token-efficient, measuring traffic by tokens unfairly disadvantages their usage metrics, potentially misrepresenting their actual market share or usage intensity. The public dispute between high-level executives at OpenAI and Cursor ultimately harms users by creating uncertainty and reducing access.

Theo criticized the public bickering between AI company leadership, stating, "Fighting over the details of how this number was measured, it just kind of feels childish and unnecessary." He emphasized that such disputes create unnecessary friction and reflect poorly on the professional image of the companies involved.

While the core of the dispute might be over metrics and contractual details, the real-world consequence is reduced access for customers and a loss of trust. The unprofessional nature of this public conflict can harm user confidence and the overall perception of the AI industry.

Theo argued that all-in-one subscription models for AI tools are often unsustainable for users, stating, "You're kind of screwed because the rates you can get from subscribing to the providers directly are just so much better."

Direct subscriptions to AI providers are almost always more cost-effective than using an intermediary. This is because AI labs like OpenAI operate with massive margins, allowing them to heavily subsidize their own services and offer pricing that third-party tools cannot match.

The economic reality makes it nearly impossible for smaller, third-party providers to compete on price with the labs themselves. This creates a significant challenge for all-in-one platforms that aim to consolidate various AI models under a single subscription.

Consequently, users are often better off subscribing directly to the individual AI services they need, rather than relying on an aggregated platform.