Meta releases Muse Spark Open, an 816B-parameter open-weight coding model

AICline announces that Muse Spark Open is available in its coding tool. The post says the model is optimized to run on Apple M5 Ultra Mac Studios with the latest MLX framework, developed in collaboration with Apple. Meta describes the open-weight model as 816B total and 42B active parameters, with a 1M token context window and native support for text, image, video, and audio.

AIMeta is targeting long coding tasks with open-weight models, which the post says makes integration a sharper competitive battleground. It adds that closed coding vendors face pressure to sell the workflow, not model access alone.

AIMeta releases Muse Spark Open, an open-weight model with 816 billion parameters (42 billion active) and up to one million tokens of context. Meta says it leads several benchmark categories against Qwen 3.8, GLM 5.3, and Kimi K3, but trails some competitors on coding, and it still falls short of Meta's paid Muse Wonder Full models. The weights are on Hugging Face under a free license that requires the label "Built with Muse Spark Open"; providers with more than $20 million in annual revenue offering it as an API need a separate agreement with Meta.

AIMeta has released Muse Spark Open, which NVIDIA AI describes as the most capable US open-weight model to date. The model has 816B total parameters with 42B active, a 1M-token context window, and native support for text, image, video, and audio. Meta also published NVFP4 checkpoints optimized for Blackwell Ultra, which NVIDIA says deliver over 50K tokens/sec/GPU.

AIArena (@arena) says users can test Muse Spark Open in Battle Mode and Agent Mode at arena.ai. The post gives no further details on the model's specifications or availability.

AIMeta announces Muse Spark Open, an open-weight model with 816 billion total parameters that it says runs on two Apple M5 Ultra Mac Studios using the MLX framework. Meta and Apple worked together on the launch, and Apple published a technical post about MLX and the model.

AIAlexandr Wang announces Muse Spark Open, which he calls the most capable US open-weight model to date. The model has 816B total parameters, 42B active, a 1M-token context window, and native multimodal support, built for coding and long-horizon agentic work. The post itself is a brief reaction, so no further claims are stated.

AIOpenRouter says Muse Spark Open is available on its platform from day 0, with Baseten, Fireworks AI, Nebius, and Together AI as launch partners. The post links to a dedicated OpenRouter model page for Muse Spark Open.

AIOpenRouter says Meta's Muse Spark Open is now available on its platform. Meta describes it as the most capable US open-weight model to date, built for coding and agentic work. The model has 816B total and 42B active parameters, a 1M token context window, and accepts text, image, video, audio, and PDF input. OpenRouter lists pricing at $1.25 per million input tokens and $4.25 per million output tokens.

AIMeta launches Muse Spark Open, a coding-optimized mixture-of-experts model with 816B total parameters, 42B active parameters, and a 1M token context window. The model was trained on NVIDIA GB300 NVL72 Blackwell Ultra systems for agentic coding workflows. The source says it offers higher first-attempt accuracy, reliable tool calling, and deep codebase understanding across repositories.

AIMeta's Muse Spark Open is an open-weight model that the OpenRouter listing describes as optimized for coding. The listing says it was trained for agentic coding workflows, with higher first-attempt accuracy, reliable tool calling, and deep codebase understanding. The feed excerpt ends before giving benchmarks or availability details.

AIMeta is releasing the weights for Muse Spark Open on Hugging Face under the Muse Model License. The model has 816B total parameters, 42B active, and a 1M-token context window, and it runs on the SGLang and vLLM inference engines.

AITogether AI launches Serverless Inference for Muse Spark Open, a managed, high-throughput production path for agentic coding and multimodal workloads. The post links to the model's page on Together AI for developers to start building.

AITogether AI says Muse Spark Open handles multi-file refactors, debugging, and long-running tasks. It holds whole projects in a 1M context window, including dependencies and legacy code. The model also accepts audio, audio-video, image, and document input, and it can plan, act, observe, and self-correct over long horizons.

AIYuchen Jin says Meta has open-sourced Muse Spark Open, which he describes as roughly Muse Spark 1.2 level. He hopes Meta will follow with open weights for Muse Spark 1.3, which he says would make it the best open-source model in the world.

AIAmazon SageMaker JumpStart now offers Meta's Muse Spark Open, a coding-optimized open-weight model with a 1 million token context window. The post walks through deploying it to a SageMaker AI endpoint in the user's own AWS account, and notes the NVFP4 variant runs on ml.p5e.48xlarge while the BF16 variant needs ml.p6-b300.48xlarge.

AINathan Lambert says that if Muse Spark Open performs as its scores indicate, adoption will be strong, calling it the first US open-weight model with near-frontier agentic capabilities. He points to recent Chinese releases, GLM 5.2 and Kimi K3, as the earlier step change. The quoted post from Alexandr Wang describes Muse Spark Open as 816B total parameters with 42B active, a 1M-token context, native multimodality, and built for coding and agentic work. The attached benchmark table compares it with Chinese and other models on agentic, reasoning, coding and multimodal tasks.

AISGLang has published a cookbook for deploying Muse Spark Open on NVIDIA, AMD, and Intel hardware. The project also says it has day-0 support for Google TPUs through sglang-torchtpu, with more details to follow.

AIMerve Noyan says Meta is releasing Muse Spark Open, which she calls the most capable US open-weight model. She describes it as an agent-focused mixture-of-experts model that accepts all modalities. The model is on Hugging Face with day-0 support in transformers and llama.cpp.

AIApple and Meta announced a collaboration that lets developers run Meta's open-weight Muse Spark Open model on Apple's high-end Mac Studio desktops. Users can pair two to four Mac Studios with M5 Ultra chips to run the 816-billion-parameter model locally, and a 256GB configuration starts at $10,799. Meta says the model is the most capable open-weight model under a license allowing limited commercial use.

AIBaseten says Meta's open-weight Muse Spark Open is available day 0 on its Model APIs and Dedicated Inference. The model has 816B total parameters, 42.7B active during inference, and a 1M token context window with image, video, and audio input. Baseten says its NVFP4 checkpoint cuts weight size from about 1.6 TB to 540 GB, and a hybrid attention design reduces KV-cache memory 3.7x at 128K context.

AIMeta is releasing Muse Spark Open, which it calls the most capable US open-weight model to date, built for coding and agentic work. The model has 816B total and 42B active parameters, a 1M token context window, and natively handles text, image, video, and audio. Ollama says the model will be available on its platform shortly.

AIAI at Meta says Muse Spark Open is available today through API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The weights can also be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. The weights are released under the Muse Model License, which has limited commercial terms that let many developers use the model for free.

AIMeta says Muse Spark Open is optimized to run on two Apple M5 Ultra Mac Studios using the latest MLX framework. The model also runs on SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth, with optimizations for AMD, Apple, Intel, and NVIDIA silicon. Developers can customize it with PyTorch, including TorchTitan for fine-tuning.

AIAlexandr Wang says Muse Spark Open is available today through API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Its model weights can be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Model Catalog. The post also says the model runs on SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, with optimizations for AMD, Apple, Intel, and NVIDIA silicon, and can be customized with PyTorch.

AIAlexandr Wang says the weights for Muse Spark Open are released today, calling it the best US open-source model. He says it runs on the new Mac Studio, citing work with Apple and John Ternus. The post includes a photo of two Mac Studios running a chat and code session.

AIvLLM announces day 0 support for Muse Spark Open, an 816B-parameter open-weight mixture-of-experts model with 42B active parameters and a 1M-token context window. The model is built for long-horizon agentic coding and supports image, video and audio input, with prebuilt images for Nvidia and AMD and initial TPU support. Nightly PyPI wheels await merging of PR #61879, which usually takes about a week.

AIMeta has released Muse Spark Open, an open-weight mixture-of-experts model with about 816 billion total parameters and about 42 billion active per token. It accepts text, images, audio, and video, outputs text, and supports a context window of up to one million tokens. The Hugging Face post provides BF16 and NVFP4 checkpoints, along with inference recipes for llama.cpp, vLLM, and SGLang and a TRL fine-tuning example.