Yesterday, a friend noted that it performed even better than OpenAI-O1! I tested DeepSeek R1, and the conclusion is just two words: impressive! I was asked if DeepSeek-R1 can be used in Cline? Today, at the request of a friend, I will do a simple test. The task this time is to create an AI version of an automatic Spring Festival couplet using Deepseek’s API.
First, let’s talk about the testing conditions:
1. Latest version of RooCline
2. DeepSeek-R1 (using the openrouter interface)
If you’re not familiar with configuring RooCline DeepSeek in detail, you can refer to this: Cline+DeepSeek can’t compete with Cursor? A full video demonstration on how to create a server monitoring app using them.
To get straight to the results: I succeeded on the first try in about 10 minutes, here’s the effect image.
However, the second attempt failed. Overall, it wasn’t a complete failure; it could definitely be finished, but I lost patience after 18 minutes. Here’s the demonstration video:(As usual, the video demonstration is unedited, with no lines of code written.)
Summary of a few shortcomings:
First, the model itself has some capability deficiencies. After opening the configuration options in RooCline, there will be reminders about missing functionalities:
From the image, we can see that the DeepSeek-R1 model has the following unsupported functionalities:
-
Does not support image processing – This means that the model cannot handle, generate, or analyze image-related tasks.
-
Does not support computer use – This indicates that the model cannot directly control or operate computer systems.
-
Does not support local file caching – This means that the model cannot save or cache files locally.
Additionally, when handling some relatively complex problems, reminders keep appearing. Of course, it claims these are complex tasks, but in my view, they should be quite simple since they are purely web-based and do not involve complex logic.
The third point, as can be seen from the demonstration video: DeepSeek-R1 tends to overthink. What is clearly a simple missing parenthesis leads to a complete modification. Of course, this issue may not solely be DeepSeek-R1’s fault; RooCline’s tuning may also have some problems.
Finally, there’s the speed issue. Compared to the previous DeepSeek-V3, the speed is indeed too slow. This may also be due to its relatively long thought process. I believe its usage scenario may be more suitable for overall project planning and implementation of individual functional points.
Therefore, the overall conclusion of the test is: the future is promising, and it’s just a matter of time; you can wait and see!
Conclusion: This project only created a very simple page. There will be local DeepSeek-R1 14b test content and more application scenarios later. Feel free to click on the cute little penguin below to follow us!
