Meta AI focuses on advancing AI through open research. Its Llama models, with the latest being Llama 4, are open-weight models designed to handle politically and socially contentious questions more effectively.
A 120B-parameter model trained on scientific literature and data, intended to assist in scientific tasks:contentReference[oaicite:14]{index=14}.
View details →A suite of foundation models (7B–65B parameters) trained on 1.4T tokens, with the 65B version outperforming GPT-3 on many benchmarks:contentReference[oaicite:15]{index=15}.
View details →The next-generation open-source Meta model (7B,13B,70B sizes), with 70B model reaching 68.9% MMLU and available under a permissive license:contentReference[oaicite:16]{index=16}:contentReference[oaicite:17]{index=17}.
View details →Latest Meta AI model (April 2025) with 400B parameters and 40T tokens trained, featuring a new Llama 4 license:contentReference[oaicite:18]{index=18}.
View details →Open Pretrained Transformer (175B) replicating GPT-3 architecture, released by Meta for research with an open training logbook:contentReference[oaicite:13]{index=13}.
View details →