Most companies announce big news at 9am Monday to get maximum attention. OpenAI dropped this one at 7pm on a Friday night, which tells you they wanted the fewest possible eyes on it. I found this breakdown from Matthew Berman, who spotted the blog post immediately and walked through the whole messy backstory, and I was genuinely surprised by how deep the drama goes.
Here’s the key idea. According to the creator, OpenAI notified SpaceX that it’s winding down the contract that put OpenAI models natively inside Cursor. Translation: OpenAI is pulling its models off the Cursor platform, giving up real revenue to do it.
🔍 The old way vs the new way
The old model was simple. Coding platforms like Cursor stayed neutral. They didn’t build their own models. They just plugged in OpenAI, Anthropic, and Google and let you pick. Everyone played nice.
The new reality is different. As the video explains, once SpaceX AI acquired Cursor, OpenAI got nervous. Their stated reason, quoted straight from the post, is that they can’t be confident SpaceX will stay within their terms of service, citing Elon Musk’s companies violating contracts before. Under oath this year, the video notes, Musk admitted XAI had violated OpenAI’s terms around distillation.
Why does that matter? The original poster breaks distillation down simply:
- You ask a big “teacher” model tons of questions
- You collect the question-and-answer pairs
- You train a smaller “student” model on them
That lets a company skip the expensive work of gathering and cleaning training data and leapfrog straight to a near-frontier model for a fraction of the cost. OpenAI’s fear, per the video, is that Cursor plus SpaceX compute could vacuum up all that model data and train a rival.
📌 The part that made me laugh
Here’s the contradiction the creator highlights. Anthropic banned XAI from using its models back in January. You’d expect them to hold that line. Instead, Anthropic co-founder Tom Brown posted that Cursor stays a trusted partner and they’re excited to keep supporting Claude inside Cursor at SpaceX. The person who shared this called it one of the most contradictory things he’s ever seen. Anthropic desperately needed compute, SpaceX had idle GPUs, and suddenly the beef was patched.
Meanwhile Cursor’s CEO says OpenAI models only serve about 5% of Cursor traffic. But as the video points out, tokens aren’t a proxy for revenue. Frontier models are token-efficient, so that 5% likely punches way above its weight in actual dollars.
🛠️ What this means for you if you use Cursor
The creator says not all is lost:
- You can still load your own OpenAI API key into Cursor manually
- You can use the Codex IDE extension and compatible gateways
- Claude and Grok models stay available natively
The bigger takeaway from the video is a trend worth watching. Model-agnostic platforms are risky, because you’re buying the core of your product from a competitor. So expect OpenAI, XAI, Google, and Anthropic to go fully vertical: their own models, their own harnesses, their own coding tools. Mixing multiple top models in one place may get harder.
And yes, Elon weighed in too, calling Sam Altman and Greg Brockman untrustworthy. The spite runs a decade deep, back to OpenAI’s 2015 nonprofit origins.
If you want the full timeline, the Twitter skirmishes, and the acquisition logic, watch the video. It’s a great breakdown of where AI coding tools are heading.