Introducing Meta Llama 3: The most capable openly available LLM to date

web

Credibility Rating

4/5

High(4)

High quality. Established institution or organization with editorial oversight and accountability.

Rating inherited from publication venue: Meta AI

Official Meta announcement of Llama 3; relevant to AI safety discussions around open-weight frontier models, dual-use risks, and the safety measures (or lack thereof) accompanying powerful openly available systems. The original tags referencing disinformation appear misassigned.

Metadata

Importance: 62/100blog postprimary source

Summary

Meta announces Llama 3, their most capable openly available large language model family, featuring 8B and 70B parameter models with improved reasoning, coding, and instruction-following capabilities. The release includes details on training data, architecture improvements, and safety measures implemented before public release. Llama 3 represents a significant milestone in open-weight frontier model development.

Key Points

•Releases 8B and 70B parameter models pretrained and instruction-tuned, with a 400B+ model still in training at time of announcement
•Trained on over 15 trillion tokens from publicly available sources, with improved data quality filtering and curation
•Introduces a new 128K token vocabulary tokenizer and grouped query attention (GQA) for improved efficiency
•Benchmarks show competitive performance with leading closed models like GPT-3.5 and approaches GPT-4 on several tasks
•Includes Meta's safety tooling (Llama Guard 2, Code Shield, CyberSec Eval) and responsible use guidelines alongside the release

Cited by 3 pages

Page	Type	Quality
AI Scaling Laws	Concept	92.0
Meta AI (FAIR)	Organization	51.0
AI Disinformation	Risk	54.0

Cached Content Preview

HTTP 200Fetched Mar 20, 202637 KB

[Go up one level](https://ai.meta.com/blog/meta-llama-3/# "Go up one level") [![Meta](https://scontent-lax3-1.xx.fbcdn.net/v/t39.8562-6/252294889_575082167077436_6034106545912333281_n.svg/meta-logo-primary_standardsize.svg?_nc_cat=108&ccb=1-7&_nc_sid=e280be&_nc_ohc=7TgIhmzWV40Q7kNvwHWWUGP&_nc_oc=AdrFmyL1nczVbuCAUzFM6Cbc8wCKMEil8lLj4d83Uvl_D91wDW_GVax0hhV2R-EHYjk&_nc_zt=14&_nc_ht=scontent-lax3-1.xx&_nc_gid=wtwdgbhbg69D2GUDupmp1A&_nc_ss=8&oh=00_AfzPcP_K2jc6Lvnpr0QMO8s_UYgjQSQhSDDhW8EjR_h6Wg&oe=69C1DF79)](https://ai.meta.com/)

- [Meta AI](https://ai.meta.com/blog/meta-llama-3/#)

- [AI Research](https://ai.meta.com/blog/meta-llama-3/#)

- [The Latest](https://ai.meta.com/blog/)

- [About](https://ai.meta.com/blog/meta-llama-3/#)

- [Get Llama](https://www.llama.com/?utm_source=ai_meta_site&utm_medium=web&utm_content=AI_nav&utm_campaign=09252025_moment)


- [Try Meta AI](https://www.meta.ai/?utm_source=ai_meta_site&utm_medium=web&utm_content=AI_nav&utm_campaign=02022026_moment)
- [Toggle site search](https://ai.meta.com/blog/meta-llama-3/# "Toggle site search")


[Close submenu](https://ai.meta.com/blog/meta-llama-3/# "Close submenu") [Main menu](https://ai.meta.com/blog/meta-llama-3/# "Main menu")

[BACK](https://ai.meta.com/blog/meta-llama-3/# "Go up one level")

- [Explore Meta AI](https://ai.meta.com/meta-ai/)
- [Get Meta AI](https://ai.meta.com/get-meta-ai/)
- [Meta AI Studio](https://ai.meta.com/ai-studio/)

- [Overview](https://ai.meta.com/research/)
- [Projects](https://ai.meta.com/research/#projects)
- [Research Areas](https://ai.meta.com/research/#research-areas)
- [People](https://ai.meta.com/results/?content_types[0]=person)

- [Overview](https://ai.meta.com/about/)
- [Open Source](https://ai.meta.com/opensourceai/)
- [Careers](https://www.metacareers.com/)

Clear

- Clear

- [Meta AI\\
\\
>](https://ai.meta.com/blog/meta-llama-3/#)

- [AI Research\\
\\
>](https://ai.meta.com/blog/meta-llama-3/#)

- [The Latest](https://ai.meta.com/blog/)
- [About\\
\\
>](https://ai.meta.com/blog/meta-llama-3/#)

- [Get Llama](https://www.llama.com/?utm_source=ai_meta_site&utm_medium=web&utm_content=AI_nav&utm_campaign=09252025_moment)

[Try Meta AI](https://www.meta.ai/?utm_source=ai_meta_site&utm_medium=web&utm_content=AI_nav&utm_campaign=02022026_moment)

Large Language Model

# Introducing Meta Llama 3: The most capable openly available LLM to date

April 18, 2024

## Takeaways:

RECOMMENDED READS

- [5 Steps to Getting Started with Llama 2](https://ai.meta.com/blog/5-steps-to-getting-started-with-llama-2/)
- [The Llama Ecosystem: Past, Present, and Future](https://ai.meta.com/blog/llama-2-updates-connect-2023/)
- [Introducing Code Llama, a state-of-the-art large language model for coding](https://ai.meta.com/blog/code-llama-large-language-model-coding/)
- [Meta and Microsoft Introduce the Next Generation of Llama](https://ai.meta.com/blog/llama-2/)

- Today, we’re introducing Meta Llama 3, the next generation of our state-of-the-art open source larg

... (truncated, 37 KB total)

Resource ID: f9616f30e8f51cb0 | Stable ID: MmI1NzhkNz