BaseRT is a native Metal Inference runtime for large language models (LLMs) developed by Base Compute (Melbourne/Berlin) specifically for Apple Silicon devices, without relying on intermediate layers such as MLX, PyTorch, or CoreML. The product consists of a CLI (basert), a proprietary model format (.base), an OpenAI-compatible server API, and language bindings for Python, Node, Rust, and Swift. According to the manufacturer and accompanying research papers, BaseRT achieves the highest measured Inference throughput on Apple M-series chips among all tested local runtimes, including faster performance than llama.cpp and MLX. The core repository is Open Source under Apache 2.0 license, while the precompiled engine binary is subject to a separate proprietary license.

Preuves (2)