Open R1: Update #4

This week, a new model from DeepSeek silently landed on the Hub. It’s an updated version of DeepSeek-V3, the base model underlying the R1 reasoning model. There isn’t much information shared yet on this new model, but we do know a few things!

The model has the same architecture as the original DeepSeek-V3 and now also comes with an MIT license, while the previous V3 model had a custom model license. The focus of this model release was on improving the instruction following as well as code and math capabilities. Let’s have a look!

The DeepSeek team has evaluated the model on a range of math and coding tasks and we can see the model’s strong capabilities compared to other frontier models:

Given the naming and architecture it is fairly safe to assume that the new model is based on the previous V3 model and trained on top of it. There are two possible areas how they improved the models:

So downloading the weights is safe, and upon code review so is executing the modeling code. This means you can run the DeepSeek model locally without the risk of backdoors or malicious code execution.

Clearly, the model plays in the top league: often on par with GPT-4.5 and generally stronger than Claude-Sonnet-3.7.

Specifically, in the model card the DeepSeek mentions targeted improvements in the following areas:

So the question might pop-up: how did they actually do this? Let’s speculate a bit!