What is a private LLM, and does your company need one?
"Private LLM" means different things to different vendors. Here's what the options really are, and how to tell which one you need.
Short answer: a private LLM is a language model that runs where you control it, such as your own cloud account or your own servers, so your data never goes to a shared service. Most companies don't need one: the business plans of ChatGPT and Claude already keep business data out of model training by default. You need a private model when regulation, contracts or the sensitivity of the data rule out any third-party service, or when you want a model trained on your own work.
The four levels of "private"
| Option | Where it runs | Who sees the data | Fits |
|---|---|---|---|
| Business plan (ChatGPT, Claude) | Provider's cloud | Provider, under business terms | Most companies |
| Provider via your cloud | Your AWS, Azure or Google Cloud account | Your cloud provider, under your existing agreement | Companies with cloud compliance already set up |
| Open model, self-hosted | Your cloud or data centre | Only you | Regulated data, strict contracts |
| Your own trained model | Wherever you choose | Only you | A core task where you want an edge |
When you actually need one
- Your contracts or regulators forbid sending certain data to any third party.
- You need guaranteed behaviour that doesn't change when a provider updates its model.
- You run a high-volume task where a small self-hosted model is much cheaper per request.
- You want a model trained on your own work that competitors can't buy.
The trade-offs
Self-hosting means you own the uptime, the security patches and the upgrades. Open models that fit on modest hardware are usually behind the best frontier models on general tasks, which is why the strongest private setups are trained for a specific job. See how to build your own model.
Questions
Is ChatGPT Business private?
OpenAI says data from ChatGPT Business and Enterprise isn't used to train its models by default. It still runs on OpenAI's service, so check it against your own contracts and regulators.
What is the best LLM to run privately?
It changes every few months. Choose by testing open models on your own tasks, at the hardware size you can afford to run.