You Can Now Use DeepSeek R-1 On Cursor AI
As expected, DeepSeek R-1 and V-3 would eventually make their way to AI coding tools like Cursor AI.
As a developer who’s been using AI for the past year or so in my personal development projects, I can’t help but be extremely excited now that the DeepSeek R-1 model is now supported in Cursor AI.
The new model support was announced on X by DeepSeek.
Deepseek models are available now in Cursor! Hosted on US servers. While we’re big fans of Deepseek, Sonnet still appears to perform much better on real-world tasks. Enjoy!
The real question now is: is it as good as—or even better than—Claude 3.5 Sonnet? Let’s find out.
What is DeepSeek R-1?
DeepSeek R-1 is the newest reasoning model developed by the Chinese AI lab DeepSeek. It’s fully open-source, allowing developers to freely use, modify, and fine-tune the codebase according to their needs.
According to DeepSeek’s internal benchmarks, R-1 and o-1 models perform on par in multiple key categories, including mathematical reasoning (AIME benchmark), competitive coding (Codeforces), and complex question-answering tasks (GPQA Diamond). The performance difference between the models in these areas is often within a few percentage points.
And just a quick fun fact: this model actually just started as a side project to make use of DeepSeek engineers’ extra GPUs. A side project that’s now competing with billion-dollar language models? That’s not just impressive—it's pure genius.
If you want to learn how to install and use DeepSeek R-1 on your local PC, check out this article.
How To Enable DeepSeek In Cursor AI
If you don’t have Cursor AI installed on your system, go to cursor.com and download the installer. Make sure you are logged into the website to be able to claim your free credits.

Once istalled, open the Cursor app and go to the settings menu. Under the Models tab, you’ll notice that deepseek-r1 and deepseek-v3.

You can now close the settings page, and when you pull up the chat section (Cmd + shift + k), you should see both deepseek-r1
and deepseek-v3
added to the list of models.

One interesting characteristic of DeepSeek models is that they often start and end their responses with <think> tags.

This reminds me of how this reasoning model uses its internal dialogue to come up with its response.

Feel free to experiment with both deepseek-r1 and deepseek-v3 in Cursor AI. Use them to build and debug your code projects.
However, keep in mind that if you’re on a free-tier account, you won’t be able to access these models in Composer mode.

Cursor AI will prompt you to upgrade to a Pro or Business plan to unlock full access.
Should You Use DeepSeek R-1 On Cursor AI?
This is probably the biggest question on your mind: is DeepSeek R-1 worth using in Cursor AI?
Based on my experience, the model performs exceptionally well in understanding code and identifying bugs. However, when it comes to writing or modifying code across multiple files, it doesn’t match the performance of Claude 3.5 Sonnet.
I don’t have the benchmarks or specific examples to illustrate why I still personally prefer the Claude 3.5 Sonnet model over any other model, including DeepSeek R-1 or V-3. I just feel that Claude is more optimized for prompt understanding and code generation.
Here are some of my personal findings:
Single-file coding tasks: It’s good at concise, logical problem-solving (e.g., bug fixes, algorithm design).
Multi-file projects: It struggles to maintain context across multiple files, making it less effective for large-scale projects.
Prompt sensitivity: DeepSeek R-1 requires precise, zero-shot prompts to deliver optimal results. Claude, on the other hand, handles few-shot prompts more flexibly.
Another reason is that DeepSeek models are not available in the composer mode. You need to upgrade to a paid account to be able to use them. So if you’re only intending to test the Cursor AI features, I highly suggest sticking to the free models on a free account.
But where DeepSeek truly shines is that it's very cheap. For a typical development session (50k input + 10k output tokens):
DeepSeek-R1: $0.48
Claude 3.5: $3.15
The model’s $6 million training cost suggests a sustainable development path, potentially pressuring closed-source providers like OpenAI and Anthropic to reduce prices.
For developers willing to adapt their prompting strategies and workflow, DeepSeek-R1 offers really good value in reasoning-focused tasks despite current limitations in complex software engineering.
Cursor AI Pricing
One of the things I appreciate about Cursor AI is the generous free credits. When you sign up, you get access to the Pro version, which includes the following benefits:
Pro features for two weeks
2,000 completions
50 slow premium requests
200 cursor small uses
The free credits should be enough to test out all the features and build simple web apps. However, if you’re planning on creating more complex projects with advanced features, you might want to consider upgrading to one of the paid tiers:
Pro: Priced at $20 per month, this plan includes everything in the Hobby tier but adds unlimited completions, 500 fast premium requests per month, and unlimited slow premium requests.
Business: This tier costs $40 per user per month and encompasses all Pro features, along with centralized billing, an admin usage dashboard, enforced privacy mode, and OpenAI or Anthropic zero-data retention policies.
Final Thoughts
It was expected that DeepSeek R-1 and V-3 would eventually make their way to AI coding tools like Cursor AI, especially with their open-source nature, making adoption easier.
Based on my initial experiments, the model is really good when it comes to code understanding and debugging. It also performs well in areas like logic and math, as shown by its benchmark results. However, when it comes to handling multi-file code modifications or complex prompts, it’s not quite up to the level of Claude 3.5 Sonnet. These limitations can be a dealbreaker for larger or more complicated projects that require maintaining context across multiple files.
Now, if you care about cost efficiency, DeepSeek R-1 is a very attractive option. Its pricing is significantly lower than Claude 3.5 or even GPT-4o. However, if you want the best performance and high-quality code generation, you’re better off sticking with Claude 3.5 Sonnet.
Having more options like DeepSeek R-1 is always a win for developers and even for regular users. More competition means companies like OpenAI and Anthropic will likely be under pressure to release cheaper models to stay relevant. This increased variety and affordability will only help us build better apps and push AI development further.
Have you tried DeepSeek R-1 or V-3 on Cursor AI? If yes, I’d love to know your thoughts on this brand new AI model.
Hi there, thanks for making it to the end of this post! My name is Jim, and I’m an AI enthusiast passionate about exploring the latest news, guides, and insights in the world of generative AI. If you’ve enjoyed this content and would like to support my work, consider becoming a paid subscriber. Your support means a lot!