Rapid-MLXProduct update
Rapid-MLX · v0.13.0
Rapid-MLX v0.13.0 runs supported Qwen3-Next checkpoints larger than available memory by streaming from disk, and serves Nemotron-Labs-Diffusion 3B as a standard text model. Hybrid models reuse compatible prefix caches, which the project says cuts verified repeat-turn prefill from tens of seconds to under a second. The desktop app adds a photo flow and separates Speech to Text from Text to Speech, so ordinary dictation no longer evicts the active chat model. First setup recommends models based on live memory, and switching models during an active request asks for confirmation first.
Saved . A shortened excerpt from GitHub Releases; the original publisher retains authorship.