Kimi K2.5 in Cursor

Model looks highly capable, can we get this? Now with vision support!

38 Likes

Give me that

5 Likes

Yea, not sure why this is taking so long - they’re usually Johnny on the spot with the other models… even ones that weren’t desired.

4 Likes

This is an open source model, [inference] providers need to pick it up first. This can take a while depending on the provider, and I’m not sure which one Cursor uses for the open source kind.

1 Like

What we really need is an “agent swarm” feature. Where if we spin up a detailed plan prompt than Kimi K2.5 for example, can spawn 100 parallel agents to execute. So you could theoretically build an entire app in just 10-20 minutes.

3 Likes

Would this end entire Cursor Pro subscription in just 10 minutes? The concept looks promising for super cheap models, though.

I’m afraid this would break existing Cursor agent loop in first place, especialy since Cursor introduced its own subagents recently.

But other than that - yeah, I’d like to have a go-to model cheaper than GPT 5.2 which is now overkill for a majority of tasks.

Cursor uses fireworks for some models according to their security page: https://cursor.com/security#infrastructure-security

Kimi is already available on fireworks with great speed and uptime:

Hope we get to see it! :slightly_smiling_face:

2 Likes

That was quick! Fireworks didn’t have it when I checked OpenRouter around the time I sent my message.

1 Like

Well they can introduce it first without the agent swarm and add that one later. Because the model is really good, considering the price.

1 Like

Regarding the ‘agent swarm’ - I think this is a feature specific to Kimi’s first party clients (e.g. their website) and are unlikely to be introduced to their API. Future will show. However as everybody has already mentioned, it would be great to have this model as-is.

I want this, i could never really get cursor to work with qwen via openrouter …. I dont know if this is intentional? I imagine I would eventually switch ide’s if cursor doesnt start prioritize supporting kimi / qwen as a priority.

Whats the hold up? is there any quick way to add custom one on our own with outside ref?

1 Like

I tried to add it via openrouter, it handles chat but as soon as it wanted to touch any file it crashed. Same with moon api directly.. so most likely we have to wait.

very insightful… :roll_eyes:

Bump. Need good coding models that are also cheap.

2 Likes

I’d love access too

C’mon Cursor open source FTW!

2 Likes

Give me that!!!

1 Like

There hasnt been any update on nightly since yesteday so don’t get your hopes up yet. If you want to test drive K2.5 there are limited altertive for free for a week. To maybe fill the gap.