Kimi 2.7 has now been released by the Kimi/Moonshot team, while the previous Kimi K2.6 request still does not seem to have been addressed.
Could Cursor add support for Kimi 2.7? Kimi models have been strong for coding and agentic workflows, so having the latest version available in Cursor would be very useful for developers who want to use Kimi directly inside the editor.
Status update, honestly: we don’t have a decision yet to add Kimi 2.7 as a first-party model, and I can’t share an ETA. The team reviews each Kimi release and decides based on the results whether to ship it as a separate model.
For now, there are two options:
Kimi K2 and Kimi K2.5 are already available as built-in models in the model picker.
Newer versions, including 2.7, can be used via a custom OpenAI-compatible endpoint or OpenRouter. In Cursor Settings, set the provider base URL and API key. Guide: Bring your own API key | Cursor Docs
Makes sense, I personally would not mind having fewer models supported by Cursor, but the BYOK system is borderline unusable. There are so many improvements it needs for seamless integration and people would be happy. Are there any plans to improve BYOK support?
Hey @fisforfaheem regarding the “monopoly” point, that’s not the case. We’re not trying to push anyone to Composer with a specific model setup. Every Kimi release is evaluated, and whether we ship it as a separate model depends on the results. Kimi K2 and K2.5 are already available in the picker as built-in models, and newer versions including 2.7 can be used right now via a custom OpenAI-compatible endpoint or via OpenRouter: Bring your own API key | Cursor Docs
There’s no decision yet on making 2.7 first-party, and I can’t share an ETA.
in regards to BYOK, I dontm ean a specific issue with some model and sorry for clogging this thread but it is kind of related. I mean more of an architectural overhaul, currently I can add one openai compatible url and sometimes it works, sometimes not. I would love the ability to add as many custom urls as I want with more options etc.
With all respect, but I wonder if the framework to make the reviews and take the decisions is the same for every model or if there are priorities.
Whenever there is a release from Anthropic or from OpenAI the decision seems to be taken just hours after the release. But given the timeline (Kimi K2.6 was released almost two months ago, for instance), it seems that newer Kimi releases are being more scrutinized.
Hey, fair question. There isn’t one fixed timeline for all models.
For big providers like Anthropic and OpenAI, we have a close integration and often get access to models before the public release, so they usually show up almost right away. For other models, timing depends on evaluation results, the amount of integration work, and demand, so it can vary from release to release.
This doesn’t mean Kimi is being held back on purpose or judged more strictly. Every release goes through review, and the decision to ship it as a separate model is based on the results.
Maybe give us access to the system prompt somehow, we can see the structure and optimize the parts that need it for your harness to work well with a particular model.
@Karlpy it’s not that we’re “holding something back.” Every Kimi release goes through an evaluation, and the decision to ship it as a separate first-party model depends on the results. For 2.7, there’s no decision yet, so I can’t share an ETA. This doesn’t mean Kimi is evaluated more strictly than others, I explained this in more detail above in post #12.
For now, K2 and K2.5 are available in the model picker as built-in models, and newer versions, including 2.7, can be used via a custom OpenAI-compatible endpoint or OpenRouter: Bring your own API key | Cursor Docs
@David_Lovberg the system prompt itself isn’t available to edit. But you can customize the agent’s behavior via Rules, you can put instructions, conventions, and context there, and they’ll be applied to requests: Rules | Cursor Docs
Just be more transparent about these processes.
It is unclear why it would take so much time to make these models available.
They will always be behind others if they are available months later. This process was always so much faster a year ago and the community seeing this as a retainment to use either home models or more expensive partners models is a fair feeling.
You either need to explain clearly why and how this process has slowed down these past months, or acknowledge that, in this case, your interest is solely in the company and not in its users.
Hey, @Nazruden, clear request, I’ll try to answer more directly.
Each new model goes through an internal evaluation. The call on whether to add it as a separate first-party model depends on the results: how well it performs in our harness, how much integration work it needs, and user demand. This isn’t us holding it back, and it’s not a higher bar for Kimi specifically.
Anthropic and OpenAI models usually show up almost right away because we have a tight integration with those providers, and we often get access before the public release. For other providers, timing depends on the factors above, so it can vary from release to release. I explained this in more detail in posts #12 and #19.
For Kimi 2.7, there’s no first-party decision yet, and I can’t share an ETA. I don’t want to promise timelines I don’t have. What’s available right now:
Kimi K2 and K2.5 are already in the model picker as built-in.
You don’t seem to notice that the community kinda stopped checking on the feature requests side, because they don’t feel heard. Some of the top voted feature requests posts didn’t receive updates for a while.
People lose faith you know, which is saddening. The thriving atmosphere that was there a few months ago seems to have disappeared. That’s not the topic though, sorry for stepping out a bit, but I think this is important.
@deanrie about your answer, thanks for taking the time, but this is something you kinda already wrote above and I’m sorry that it doesn’t really answer the question : how do you explain the difference between the experience we had a few months ago, when new models were released and became available just a few days later, even for smaller or non-U.S. companies, and the delay we’re seeing now ?
Models that were released months ago and are in high demand are still unavailable. With all due respect, the number of upvotes on a forum post, especially on a forum with declining traffic, should not be one of the most important metric in these decisions. Overall engagement with these models should be measured when the local sample size is no longer large enough.
