Claude's much awaited 4 series of models releas...oh. Their models are so good for coding that they can go with minor semantic versioning, ha.
During my initial testing of Claude 3.7 Sonnet, I found that it successfully passed the vibe check.
It follows the instructions well and is faster than 3.5 Sonnet and 3.6 (no, I am not slapping "new" on it) Sonnet. Something that I both cherished about xAI's Grok 3 model.
Prompting Claude Models
It has been nearly a year of working with Claude models since I explored 3 Opus, and mostly stuck with it for iOS development. I am glad I followed their prompting guide, which I think was beneficial at that time to steer the model towards the desired output.
I feel the model has improved enough to become more smarter and you do not need to explicitly prompt as much as you used to before. Still, here are some general guidelines to follow:
Generating Prompts
Also, I love Anthropic's prompt generation option that can easily spit out prompt personalized for the model itself. You can find it in their console:
27th February 2025
UPDATE: Since the launch, the model as degraded and the below examples are not effective anymore
Refactoring
I am working on an MLX Swift project and have to refactor a project to fit into another existing project. I believe that 3.7 Sonnet is capable of one-shotting this task, given the appropriate prompt and explicit files to add.
I decided to take the lazy route, duplicated the first project, and asked Claude to change the instances of the name everywhere. Back in the day, I would slog through this manually—yuck.
This time, I let the agent do its thing while I went to satisfy my late night sweet craving.
Out of 32 successive tool calls, only 1 failed because it was a binary file and could not be opened as text. This is something I have not seen any model do before.
Creating New Files
Next, I took the code structure and all the files from the other project, and asked 3.7 to integrate the specific functionality that I wanted.
Again, zero tool call failures. It was able to go through the relevant files and find the exact part to integrate while creating a new SwiftUI view.
First reaction: it spits out a LOT of code and text in one go.
Almost 1000 lines of code.
It took me a while to go through the file and I was surprised to see only 1 error, a duplicated SwiftUI loading image view, and that is forgiven.
It simulated demo code in a few places, and that is my prompting issue as I am creating a prototype and used the word "demo" in the prompt. Then, I asked it to use the actual files. No prompting engineering, just simple one line sentence.
And it did. It went ahead and added those file to the project, referenced them in the newly created view and I was able to get a prototype done while writing these notes!
Moving Forward
It is 4:00 AM of February 25th, 2025, and I have some liberty to be dramatic.
I want to liberate myself from the process of coding so I can enjoy the output. And this, 3.7 Sonnet, is a model that gets me more than a few steps forward in that direction.
These are some of my initial impressions, and I will continue updating this page as I use the model more. You can bookmark this page or follow me on Twitter for updates:
Happy Clauding!
Master AI-assisted iOS Development
Learn how to use Alex Sidebar, Cursor, Windsurf, VS Code, and other AI tools for your Swift and SwiftUI workflow.