Meta's largest open-weight model in the Llama 3.1 family, running in FP8 precision — a quantization format that reduces memory footprint while preserving most of the full-precision quality. It handles complex reasoning, multilingual tasks, and long-context work across an 128k token window. The FP8 format makes it more practical to self-host than a full BF16 version, though users should expect minor quality trade-offs compared to full precision.