Meta AI

FAIR (Facebook AI Research) was founded in 2013 under Yann LeCun. Meta consolidated FAIR, GenAI, and Reality Labs research into Meta AI, and pursued an aggressive open-weights strategy from Llama 1 (2023) onward.

Funded by Meta Platforms; 2024 capex exceeded $35B with AI as the largest line item.

• PyTorch

  Reference deep-learning framework, now under Linux Foundation.

• Llama 1 / 2 / 3 / 4

  Open-weights LLM family; defined the open ecosystem.

• Segment Anything (SAM)

  Foundation model for image segmentation.

• V-JEPA

  LeCun's joint-embedding predictive architecture for video.

• Llama frontier models
• FAIR long-horizon research
• AI for Reality Labs (AR/VR)

• Llama Guard family
• Purple Llama responsible-release toolkit
• Open weights with use policy

• Llama 2: Open Foundation and Fine-Tuned Chat Models

  arXiv · 2023

• Segment Anything

  ICCV · 2023

• PyTorch 2016

  Became the dominant research framework.

• Llama open-weights release 2023

  Catalysed the global open-source LLM ecosystem.

• Microsoft (Llama distribution)
• Hugging Face
• Nvidia

1. 2013

   FAIR founded under Yann LeCun.

2. 2016

   PyTorch released.

3. 2023

   Llama 1 and Llama 2 released.

4. 2024

   Llama 3 sets open-weights frontier.

The only major lab pursuing fully open weights at frontier scale. Strategy aims to commoditise the model layer so Meta's distribution and hardware investments dominate.