Kimi K2.5 in Cursor

Ideas Feature Requests
44

Are you concerned about security issues, I think the kimi code extension is still in technical preview?

Also does it allow the agent swarm feature?

45

I canā€™t imagine how this would work for regular development. Imagine the merge conflicts if 100 devs work on the same app.

46

What are they waiting for to implement the model natively?

47

Theyā€™re waiting for Trumpā€™s sovereign approval at the national level.

5 Likes
51

Honestly am surprised it has not been added yet. as far as i know, this model is genuinely fun and good to use

1 Like
52

I got the KimiK2.5 running on CURSOR using the ā€œoverride Open ai urlā€ feature.
The inference url: https://api.fireworks.ai/inference/v1 Fireworks is blazingly fast, but you will soon hit rate limits.

54

Bruh they integrated Opus 4.6 right away. Weā€™re still left waiting for Kimi K2.5

7 Likes
55

@Colin when can we expect integration? Whatā€™s up with the holdup?

56

ė¹Øė¦¬ ģ¶”ź°€ķ•“ģ£¼ģ„øģš”

58

Kimi K2.5 is absurdly cheap. If users realize a near-free model handles most coding tasks, the justification for $60ā€“$200/month tiers falls apart. Those tiers exist because serving expensive models like Opus costs real money. Iā€™d happily keep paying $20/month and use K2.5, thatā€™s actually more margin for Cursor. But the risk for them is that users stop upgrading. Classic cannibalization dilemma. Would love transparency from the Cursor team on whether this is on the roadmap.

4 Likes
59

Thatā€™s spot on analysis. I think what they should do instead is just promote agent swarms if they still wanna justify value for power users, so users still can burn tokens and demand for highter tiers, rather than artificially limit access to models

2 Likes
61

I donā€™t think they can charge whatever they want, because open-source software will provide better and cheaper alternatives once it has caught up. Cursor is obviously playing the money game, but not for long.

3 Likes
67

I have seen some videos and benchmarks of Kimi K2.5. I would love to try this little guy in Cursor and see if it can solve token cost problems.

4 Likes
69

@deanrie please add this.

70

But does that really matter? With long running agents, bugbot, and multiple runners, users will use the credits ā€œdoing moreā€

Another point to consider is: $200 users often consume $600 in tokens. Whilst many $20/60 users on Kimi could end up only using a fraction of their allowance

I think we are more likely to see a tuned Kimi as a Composer model, maybe we will get the base at the same time. Very high TPS for x2-3 the price, which will pull users away from the base model

2 Likes
71

composer 1.5 out with price 3.5 input and 17.5 output, I donā€™t think we will see kimi 2.5 soon.

3 Likes
72

Still waiting on Kimi K2.5, while Opus 4.6 was added instantly.
Iā€™ve seen a few sources that are saying thereā€™s some issues with Kimiā€™s API that make tool calling crash and most likely problematic in the context of Cursor.

I would at least appreciate some of transparency from Cursor on exactly why K2.5 has not been added.

It certainly feels like Cursor does not want to add these token efficient models to prevent cannibalization of higher plans. While Iā€™m not sure if this is true, transparency from Cursor would easily resolve this speculation.

Even just a response to this thread would be better than nothing!

17 Likes
76

this is an old model, not the new K2.5

77

Ah, why is the price not listed for K2.0 though..

78

Hi @Colin @condor ! Could you guys be a bit more transparent about adding Kimi & the other low cost, high performance models? Iā€™ve already started using kimi code and other alternatives but would love to go back to cursor if they were added

2 Likes