https://www.sharpai.org/benchmark/ The MLX part is what we've done with SwiftLM, the local result is still being verified more details are on-going.