I was able to run the full version of the just-released DeepSeek R1-0528 on my M3 Ultra Mac Studio.
It worked…with 400 GB of RAM.
Here's how it went, along with some technical notes on the whole process for LM Studio and Ollama: macstories.net/notes/testing-d…
Testing DeepSeek R1-0528 on the M3 Ultra Mac Studio and Installing Local GGUF Models with Ollama on macOS
DeepSeek released an updated version of their popular R1 reasoning model (version 0528) with – according to the company – increased benchmark performance, reduced hallucinations, and native support for function calling and JSON output.Federico Viticci (MacStories)

Mathias Hasselmann
in reply to daniel:// stenberg:// • • •daniel:// stenberg://
in reply to Mathias Hasselmann • • •Generalize for any repo by jericson · Pull Request #16 · curl/stats
GitHubWilliam Salmon
in reply to daniel:// stenberg:// • • •That said I have found that while having complexity in code is bad, it's also hard to reduce and one way that some people try to reduce it is to "hide" it across many simple looking functions.
I once got my pen and paper out and found that to get from one function at the high level to the low level equivalent the path was through 5 lays of inheritance and 3 different branches
William Salmon
in reply to William Salmon • • •daniel:// stenberg://
in reply to William Salmon • • •poleguy looking for lost tools
in reply to daniel:// stenberg:// • • •My insight from that experience is that what really matters is that the complexity of the code is as low as possible considering the complexity of the problem. But the complexity metrics do not consider the problem complexity. I would rather see all the code for a complex problem in one place.
sammypanda
in reply to daniel:// stenberg:// • • •Maage
in reply to daniel:// stenberg:// • • •