DeepSeek R1 is Insane

All of my articles are 100% free to read. Non-members can read for free here.

This week, Chinese startup DeepSeek announced the release of their newest Open Source model - R1, and dropped one sentence that nobody was expecting:

Performance on par with OpenAI-o1

Before we dive in, here’s the tweet that started all the hullabaloo:

First things first, I think we can all agree, OpenAI o1 is an incredibly powerful model, and after releasing their $200/mo Pro Plan people have been using it to do some pretty interesting stuff. In fact, people are using the Pro Plan so much that Sam Altman announced that the company is actually losing money on the plan, an unexpected but not altogether bad outcome.

Now, an Open Source model, backed by an MIT License that allows people to use it commercially with pretty much no strings attached, is putting the pressure on o1, a model that really was in a league of its own just last week.

Okay, so let’s dive into the good stuff, here’s a deeper dive into some benchmarking:

As you can see from the last line, DeepSeek-R1-Distill-Llama-70B beats o1-mini in every test…