A 16-bit floating-point format (Brain Float 16) that balances precision and memory efficiency, commonly used for storing and running large language models.