Skip to content

Pick your style: Denkbaar

Book a conversation

Articles

History

Llama 3.1 405B: an open model at frontier size

·

Meta releases Llama 3.1, with a model of 405 billion parameters and improved 8B and 70B models.

Two developers examine a large model module beside a server rack.
Llama 3.1 makes a very large model available to run yourself. Illustration: Denkbaar.

On 23 July 2024, Meta releases Llama 3.1. The largest model has 405 billion parameters.

What happened

  • Meta calls Llama 3.1 405B the largest and most capable openly available foundation model at that time.
  • The 8B and 70B models get a new version: multilingual, with a context window of 128,000 tokens.
  • Meta reports more than 300 million downloads of all Llama versions together.
  • Meta also names new uses: with the large model, you make data to train smaller models (distillation).
A large module supplies cards to two smaller modules.
The large model can create data to train smaller models. Illustration: Denkbaar.

Why it mattered

For the first time, you can download a model of frontier size yourself. That changes the choice between renting from a vendor and running it yourself.

Hands open a licence folder beside a model module and a server.
Read the licence before choosing to run the model yourself. Illustration: Denkbaar.

What you can do with it now

Read the licence. One example: if your product has more than 700 million monthly active users, you must ask Meta for a separate licence. For most companies that does not matter, but it shows that this is not a standard open source licence.

Written afterwards on 12 November 2026, from the sources below.

Sources

  1. Meta: Introducing Llama 3.1: Our most capable models to date
  2. Meta: Llama 3.1 Community License Agreement

Topics: open-sourcemodelsmeta