DeepSeek Has Arrived In Cursor — But Is It Better Than Claude ...
DeepSeek’s two models, DeepSeek V3 and R1 have been freshly added to Cursor. Many developers at the moment are using Claude 3.5 Sonnet (the latest version, claude-3-5-sonnet-20241022) as their primary LLM in Cursor (including myself), so I wanted to test out these models and see how they compare in practice.
If you haven’t heard of them yet, DeepSeek is a Chinese AI startup that’s been blowing up in the news this week as it’s just open-sourced its DeepSeek R1 model, which is boasting competitive metrics in comparison with Open AI’s o1, which is of course no small feat. Its coding-related benchmarks show that it should be better than both Claude 3.5 Sonnet and GPT-4o most of the time, which is promising. Cursor as always is quick to add new models, so let’s dive into a practical comparison!
Testing Out Different Models
We’ll be testing out chat and composer for each model and comparing the results to Claude 3.5 Sonnet. Note that agent mode is only available for Anthropic models and GPT-4o right now, so unfortunately we won’t be able to test out agent mode at the moment.
The chat task is to tell me how I would properly add a server action to a dialog in a Next.js app. This is the exact prompt:
Coding tutorials and news. The developer homepage gitconnected.com && skilled.dev && levelup.devSenior software developer at Lazer. I love writing about all things dev. linkedin.com/in/jksaunders to connect!
