Meta Llama 3 Review — 8B and 70B Open-Weight Models That Redefined the Tier

Meta Llama 3 launched April 18, 2024 with 8B and 70B models that outperformed every previous open-weight model at their respective parameter scales. A 128K-token vocabulary, 15 trillion training tokens, and Grouped-Query Attention across both sizes — at 8,192-token context. This review covers the architecture, benchmarks, training data, deployment options, and the three-month gap before Llama 3.1 extended context to 128K and added the 405B.