← Internal Conversations

What’s going on with Chinese open-source models

2 September 2025

So, Martin Casado, a partner at a16z, said up to 80% of AI startups are now using Chinese open-source models. Not OpenAI’s models. And not Anthropic’s models.

We’re talking about models like DeepSeek, Qwen, Kimi, GLM and others.

Startups are using Chinese open-source models

So what?

Early-stage startups are broke. Of course they’re using Chinese open-source models because they’re cheap!

Valid point. But there are even more reasons to use these models.

Really

You can self-host them, which means you fully control your data. Plus, you can fine-tune them for specific use cases and modify their parameters.

But open-source models must be far behind closed-source ones 🤭

Not really!

These Chinese open-source models are getting better, and it’s happening very fast 💨

On average, these models lag just 6 months behind the most advanced closed-source models from OpenAI, Google, Anthropic and X.

In fact, after the release of OpenAI’s o1, the open-source community caught up in only 4 months.

Oh

That is something!

But why do they open-source them 🤌

Some people say they are building open-source models so they can sell the hardware needed to run them.

As you know, Qwen, a family of open-source models, is developed by Alibaba.

Recently, Alibaba announced that it is actively developing and using its own AI chips.

It makes sense to me! China is playing the long game. As usual.

I see

OK, calm down!

Looks like, regardless of their origin or the reasons behind them, open-source models will drive the future of AI.

Let’s buy a GPU!

No.

...