Is Cursor Losing the Plot? Concerning Patterns Regarding Reluctance to Implement New Frontier Models Proactively

Ideas Feedback
1

I need to share something that’s starting to really weigh on me, and I suspect I’m not alone. I jumped on the Cursor Pro plan, excited to be part of what felt like the future of coding. But that excitement has quickly turned into disappointment.

The magic of Cursor was its promise: a cockpit for developers, with the best AI tools at our fingertips. But recent decisions feel like they’re breaking that promise.

We’re seeing a trend towards walled-garden integrations. Instead of getting quick access to exciting new models the community wants (where is Qwen3?) they are reluctantly added after much community outcry. When they are added, like Kimi 2, it is through a single provider that, frankly, doesn’t perform well. Why aren’t we being given the choice to use OpenRouter or other platforms? It feels less like a tool for us and more like a distribution channel for your partners.

On the Kimi 2 thread, users fought tooth and nail for this model to be added. A representative then showed up and said that there was no plan at this time to support Kimi 2. After another wave of enraged comments, we successfully forced the hand of the developers. Now with Qwen 3, the representative says they will “consider” it. It’s not “Absolutely we need to implement this today.” It is a simple “consideration”. It feels like being spit in the face.

The pricing model feel completely broken. My $60 credit, which I was happy to pay, evaporated in a handful of days. Now I’m faced with a $200 paywall. For that price, I could get 10x the value from open-source models. The math just doesn’t work anymore.

Please don’t forget who you’re building for. We aren’t asking for freebies. We are asking for value, choice, and a platform that trusts us to pick the right tools for our work. The current path of exclusive deals and restrictive pricing is pushing away your most passionate users.

I’ve decided I can’t continue with the platform in its current state and am researching other options. I just saw this tweet https://x.com/cline/status/1948088759679099130 where Qwen3-Coder is hitting 2500 t/s on Cerebras’ API with Cline. It’s very frustrating to see this level of performance achieved for free by open-source product, and Cursor users paying $30, $60 or potentially even $200 are stuck with performance from last year.

I sincerely hope you listen to the community and find your way back to the vision that we all bought into. For now, I will not be renewing my subscription unless there is a public statement made to address this pattern.

3 Likes
2

what is unfortunate is cursor is unlikely to introduce these fast apis as they have deals with companies such as fireworks

3

hi @ryunuck and thank you for your post and your concerns. Sorry that I did not get around to answer your questions as we are quite busy with new launches.

We are building Cursor for developers, therefore we regularly review feature requests for new models and have added models recently.

The process to add a model to Cursor requires certain conditions to be met, this is not an exhaustive list but perhaps minimal requirements:

  • Sufficient interest from the community in feature requests.
  • Stable model behavior and ability to be used with Agent, which requires testing.
  • Availability on one of our AI providers with sufficient capacity to serve Cursor users.

While we can not add every single new model immediately as they come out, we are definitely interested in what models our users find work well.

As noted there is a process involved before we can expose new models to users and it requires preparation.

You can always use newest models via custom API Keys and URLs. Feel free to let us know what works best.

For token usage please check my write up on how to optimize it