[SOLVED] How to use Deepseek in Cursor?

[truell20 (Michael Truell)]

Deepseek V3 and R1 are supported in 0.45 and 0.44!

You can enable them in Settings > Models > turning on deepseek-r1 or deepseek-v3. We host these models in the US.

[debian3]

But when will it support the composer?

[andresreibel (AR)]

Yes when does that happen?

[husniadil (Husni Adil Makmur)]

Will they be considered premium models or not?

[husniadil (Husni Adil Makmur)]

I noticed that it’s super fast, how many billion parameters does it use?

[condor (Condor (Tee))]

Those answers have been already supplied several times in the forum. It uses the full model, its premium but doesnt count as fast request AFAIK.

[Tomorrow (Alexander)]

How much parameters these models?

[Jeremy]

For anyone else who refuses to use the search button, its 671B parameters.

Is Cursor using FULL version of R1?

The team confirmed that the model with 671B parameters is used, not the Distill R1 version.

[Jeremiah-England]

Any idea why it doesn’t seem as good as the version you get on the deepseek website?

See

• Cursor NOT using the FULL POWER of DeepSeek R1?
• Deepseek-R1 in Cursor seems degraded compared to chat.deepseek.com

Is the explanation here most likely correct?

• Deepseek-R1 in Cursor seems degraded compared to chat.deepseek.com - #7 by normalnormie

[Don-Yin]

is it only supported in the US?

[vaisov-gemba (Sebastian Vaisov)]

Is there a way to add Deepseek with my own API key? We’re running it in Azure and would like to use it instead of Deepseek API.

[Jeremiah-England]

Cursor does not use the Deepseek API. They use Fireworks for hosting Deepseek.

See Potential concern with Deepseek R1 - #2 by danperks

[ZGarry (Garry)]

Yes, It is important!

[Valerii]

GitHub - danilofalcao/cursor-deepseek: A high-performance HTTP/2-enabled proxy server...

[Fabiandsljk (Fabiandsa)]

Please add deepseek-v3.2-thinking to our model lineup.

It currently ranks 15th on the WebDev leaderboard while being by far the cheapest option, making it the best value for web work.

Cost of deepseek-v3.2-thinking is dirt cheap:

[Juriszakers]

where is deepseek? Went to the settings and there is nothing related toDS

[douglasg14b (Doug G)]

Yeah, I don’t see deepseek in the models, and deep-seek now just got even cheaper.

[Piotr_F]

Any chance to make Cursor compatible with DeepSeek?

So far, these two cannot work together, and DeepSeek is the best value for money offer on the market.

[deanrie]

Hey, thanks for the request. Right now, DeepSeek isn’t in Cursor’s native model list. V3 and R1 used to be available, but you can’t find them in Settings > Models anymore.

You can technically connect DeepSeek using your own API key via Add Custom Model with an OpenAI-compatible endpoint, but there’s a catch. DeepSeek V4’s thinking mode requires returning reasoning_content after tool calls, and that can cause errors in the agent flow. Details and discussion here: Compatibility with DeepSeek model's design to return reasoning content after tool calls?

If you want native support back, the best move is to vote and leave a comment on the current feature request so the signal is in one place: Pls Add DeepSeek V3.2 There’s no ETA yet.