LM Studio 0.4.13 adds MLX gains

Element Labs released LM Studio 0.4.13 on May 13 with an MLX-engine update aimed at faster local vision inference on Apple Silicon Macs. The company’s changelog says build 1 upgrades to mlx-engine v1.8.1, adds parallel predictions for vision-capable models, and is “recommended for all users.” ToolMintX highlighted the release on May 19, focusing on Apple Silicon use cases for local vision models. The two posts together point to a narrow but concrete update: LM Studio is tuning image-capable model performance on Macs rather than announcing a broader product redesign. ### Which part of LM Studio changed in 0.4.13? LM Studio’s May 13 changelog says the main change in version 0.4.13 is the move to mlx-engine v1.8.1. The release notes say that engine update “significantly improves performance” and adds parallel predictions for vision-capable models, while the rest of the release is limited to a pasted-newline bug fix, bug fixes and security hardening. (lmstudio.ai) GitHub’s repository for lmstudio-ai/mlx-engine describes it as LM Studio’s Apple MLX engine and says LM Studio for Mac has shipped with mlx-engine bundled since version 0.3.4. The repository also says the engine is built with Apple’s MLX tooling and mlx-vlm for vision inference, which helps explain why this update is framed around Apple Silicon and vision models rather than Windows or Linux changes. (lmstudio.ai) ### Which models does the update call out by name? The official release notes name Qwen 3.5, Qwen 3.6 and Gemma 4 as examples of the vision-capable models affected by the update. ToolMintX’s May 19 post also points to Qwen and Gemma workflows when describing the gains for local vision use on Macs. ToolMintX says the changes are relevant for users running image understanding workloads locally on Apple Silicon, including multi-image prediction flows. (github.com) The official LM Studio notes do not publish benchmark tables in the changelog, so the company has confirmed the direction of the gains and the model families involved, but not a model-by-model speed chart in that release note. (lmstudio.ai) ### What does “parallel predictions” mean here? The release notes say 0.4.13 adds parallel predictions for vision-capable models, which indicates LM Studio can process more than one prediction path at a time for supported image models. ToolMintX described that as better handling for local vision workloads involving multiple images on Apple Silicon Macs. That reading is an inference from the wording in both posts, not a separate benchmark claim from Element Labs. (toolmintx.in) LM Studio had already been adding parallelism elsewhere in the product. Its 0.4.0 blog post in January introduced parallel requests with continuous batching, and third-party coverage of later 0.4 updates noted MLX parallel request improvements on Mac. Version 0.4.13 extends that pattern specifically to vision-capable MLX model inference. ### Why is this centered on Apple Silicon Macs? Apple’s MLX framework is designed for machine learning on Apple silicon, and LM Studio uses that stack for Mac inference. (lmstudio.ai) LM Studio’s homepage and download pages position the app as a way to run models such as Gemma and Qwen locally, while the MLX project describes its tooling as built for Apple silicon. ToolMintX’s write-up specifically referenced M3 and M4 chips in discussing multi-image prediction gains, but the official LM Studio changelog itself does not single out those chip generations by name. (lmstudio.ai) What Element Labs has formally published is narrower: MLX-engine v1.8.1 improves performance for Apple Silicon vision-capable models and adds parallel predictions. (lmstudio.ai) ### Where can users verify the update now? LM Studio’s changelog page lists version 0.4.13 as published on May 13, 2026, and marks the update as recommended for all users. The company’s main site links to the changelog, download page and model hub, while ToolMintX’s May 19 article gives a secondary walkthrough focused on Apple Silicon vision use cases. (lmstudio.ai) (toolmintx.in)