TLDR
Meta (META) shares closed at $736.72, then dipped further to $735.02 after hours.
Muse Spark Open offers 816 billion parameters and a 1M-token context window.
Meta’s new AI model can run locally using two powerful M5 Ultra Mac Studio computers.
Developers can access Muse Spark Open through leading API and cloud services.
Meta expands its open-weight AI lineup following August’s Muse Glimmer launch.
Meta Platforms shares fell Thursday, October 15, as the company released Muse Spark Open for coding and automated tasks. Meta (META) stock closed at $736.72, down 0.11%, then slipped 0.23% to $735.02 after hours. The launch expanded access to advanced AI tools, while shares recorded losses in both sessions.
Meta Platforms, Inc., META
Meta Releases Muse Spark Open for Coding
Meta introduced Muse Spark Open as an open-weight model for software development and multi-step reasoning. The system has 816 billion parameters, with 42 billion active during processing. It also handles context windows reaching one million tokens, supporting lengthy, complex coding assignments.
The model processes multiple information formats and handles programming tasks that require several decisions. Meta said the system supports long projects and difficult software engineering work. Company evaluations also compared its performance against other open-weight models available to developers.
Open-weight releases let developers download parameters and customize systems for specific operations. Businesses can therefore control deployment and manage information within their preferred environments. Meta released a smaller model, Muse Glimmer, in August before introducing this larger system.
Mac Studio Deployment Brings Hardware Demands
Meta designed Muse Spark Open to run on two Apple Mac Studios with M5 Ultra chips. The setup uses Apple’s MLX framework to manage machine learning operations locally. However, this configuration requires substantial hardware resources, including about one terabyte of memory.
The company tested Muse Spark Open alongside Muse Code, its terminal assistant for development work. One exercise involved updating a Mastodon client through software fixes and new features. These demonstrations illustrated how the system manages connected tasks without repeated manual instructions.
Meta also tested the model with DaVinci Resolve, a professional video editing application. The system reviewed footage and cropped selected sections during the demonstration. That work showed its potential for desktop tasks outside conventional programming assignments.
Meta Expands Access Through APIs and Cloud Services
Meta made the model accessible through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Developers can download weights from Hugging Face, Amazon SageMaker AI, Google Cloud, and Microsoft Foundry. These services provide alternatives to installing an expensive local hardware setup.
Compatible inference engines include vLLM, SGLang, TensorRT-LLM, llama.cpp, and Hugging Face Transformers. The new model also supports AMD, Apple, Intel, and Nvidia hardware for wider deployment. Developers can customize the system using PyTorch for specialized applications.
Open-weight distribution can reduce dependence on proprietary providers and increase deployment flexibility. However, running large models still demands technical expertise and significant computing resources. Meta’s release expands its AI offerings as Thursday’s stock action remained slightly negative.
The post Meta Platforms (META) Stock: Slips as Muse Spark Open AI Debuts appeared first on Blockonomi .
