Deepseek r1 with cursor

After setting up DeepSeek R1 with Cursor using the OpenAI API, I encountered an issue where it did not function as expected. The system returned an error message that says the “deepseek-reasoner” does not support consecutive user or assistant messages

{“error”:{“message”:“deepseek-reasoner does not support successive user or assistant messages (messages[1] and messages[2] in your input). You should interleave the user/assistant messages in the message sequence.”,“type”:“invalid_request_error”,“param”:null,“code”:“invalid_request_error”}}

1 Like

Hey I’ve got deepseek-r1 natively in Cursor now! Go check your models, you might have it there!

Thanks. I’ve already tested it, and I can confirm that r1 can be used normally. However, I did find two small problems:

1: The response speed seems to be a little slow, I don’t know why, v3 also has the same problem [if you use the OpenAI API interface to call v3, the response is very fast]

2: The previous use, the phenomenon of disconnected network connection occurred.

According to the forum reply, the cursor does not use the official API of Deepseek, but their partners, which makes the response very slow, but the accuracy and effect are not reduced. I hope it can be optimized.

We are hoping to improve the speed here, but we do use a 3rd party (Fireworks) to run this model to ensure our privacy and security agreements remain in place.

I used r1 for a day today, the effect is amazing, but the speed is too slow, so I had to switch to cline to use their official API

Okay, I understand. If possible, can you help fix this bug? The faster response of the R1 will make me enjoy using cursor even more.

Bonus news: I learned from deepseek folks that r1 doesn’t have a V3-like feature open for the time being.

R1 official API with cline in cursor?

Yeah, I am forced to use the official cline+deepseek api in the cursor because the rate

Insanely slow R1. it takes 3-4 minutes just to start thinking stage… please expand capacity for R1, now its impossible to use it normally

1 Like