The Meta company has announced the release of Llama 3, a new open-source large language model with three variants: small 8B parameter model, medium 70B parameter model, and very large 405B parameter model. The smaller models have already released weights for benchmarking, showing impressive performance compared to other models like GPT-2's 70B variant and Gemini Pro 1.5. The largest model is still in training but preliminary benchmarks suggest it compares favorably to GPT-4 Turbo. Instruct variants of both the 8B and 70B models have been released, allowing for immediate use, and fine-tuning support is available for Llama 3 8B. However, the smallest context window among these models may be a drawback compared to other large language models like GPT-4 Turbo and Claude 3.