Open models - what they are and how to use them safely

@Jensen Huang made his first X post discussing “why Open models matter” and that's amassed 50 million views in 2 days. Here’s a quick super simple breakdown of what they are, how they work, and why it's safe to use them in platforms like Cleve.

You've probably heard of ChatGPT, Claude and Gemini - and you probably use them on a day to day basis.

Beyond these models, there's a whole series of other models (often developed by Chinese labs) like GLM (from ZAI), DeepSeek and Kimi (from Moonshot).

The immediate reaction I get when I tell people about these models - is safety and security. There are concerns about how the data gets sent to China, and may be used for training & foreign intelligence by state law.

But here’s the thing, instead of keeping the models to themselves like ChatGPT, Gemini and Claude (Closed models), these labs decided to make them Open.

That means that you can even download any of these models and run it on your own computer. For free. In isolation. You can even switch off the internet and have absolutely no connection to the outside world and chat with these models.

This is a big deal. Because it means that any company can download the models for free, run them and offer them safely. This makes these models far cheaper too than the Closed models like ChatGPT, Claude and Gemini, for similar levels of intelligence.

And that’s what we did.

When you use any of the Open models (Kimi, DeepSeek, GLM) in Cleve, your data goes to a leading & trusted US-based inference provider for processing.

• The data is never used for model training (on any model in Cleve).
• We also enforce a strict Zero Data Retention (ZDR) policy and only work with inference providers that offer this. So your data doesn’t get saved in the servers.

Hence it’s one of the safest ways to try out and use these models.

Plus, because the models are lower cost, we’re able to offer more usage out of these models too at any of the pricing tiers too.

I’ve been experimenting with them a ton this last few weeks and I think they’re remarkably good models and often give better responses than the frontier models for the kind of planning, strategising & writing work I do. We also designed our own model harness to also get the most out of these models and give them much better memory.

In Cleve you can also switch, even mid-chat, between the different (Open and Closed) models too and all your context, memory and and conversations are preserved.

Try out Cleve for free, though the Open models are available on our Pro and Max plans. We also have team plans where you can subscribe once and give access to all the best models to everyone on your team.