FrontierThe story, in brief

Falcon 2: An 11B parameter pretrained language model and VLM, trained on over 5000B tokens and 11 languages

11B parameters. 5 trillion tokens. 11 languages. Falcon 2 just reset the efficiency bar for open-weight models.

Illustration of a transparent lens revealing connected networks across layers of paper.
Exploring the next frontier of AI research.AI illustration by KeyNews
The KeyNews take

Why it matters

Falcon 2 challenges the scale-first narrative by delivering competitive performance at 11B parameters across multilingual and vision tasks, directly threatening the cost assumptions of larger proprietary models.

The key facts

6 to know
  1. 11B parameter architecture

  2. 5000B+ tokens training data

  3. 11-language multilingual support

  4. Vision language model (VLM) capability

  5. Open-weight release on Hugging Face

  6. May 2024 release date

Go to the source

Hugging Face Bloghuggingface.co

Read original report
Back to today's editionMore frontier news

Keep reading

Related stories

More from Frontier