Meta Muse Spark Open goes live on Together AI

Together AI said Meta’s Muse Spark Open is live on its platform on October 15, 2026, making the open-weight model available through serverless and dedicated infrastructure on the day of launch. The company described itself as a day-zero launch partner and said early-access customers had already tested the model before release.

What Together AI said about the model

In its announcement, Together AI said Muse Spark Open is designed for long-running agentic coding workflows that keep entire codebases in context, as well as computer use, web automation and multimodal understanding of documents, images, audio and video. The company said the model has a 1M-token context window, supports long-horizon autonomy, and can plan, act, observe and self-correct over extended tasks. Together AI also said the model includes a commentary mode that states what it is doing and why before acting.

Meta’s own release, published the same day, identifies Muse Spark Open as an open-weight model with 816 billion total parameters, 42 billion active parameters per token and a 1,048,576-token context window. Meta says it accepts text, image, audio and video inputs and is aimed at code generation, review, complex software engineering, multi-step reasoning and long-running agentic tasks. Those capability claims come from Meta’s documentation and are vendor-described, not independently verified performance results.

Early access and launch partners

Together AI cited early-access feedback from Factory and Ollama. Factory’s Zimu Li said the model “consistently selected the right tool and kept its responses compact,” while Ollama’s Dong Chen said it performed well on tool calling and long context, including multi-step and parallel calls with clean structured outputs. Both companies said early access through Together AI let them evaluate the model ahead of launch and offer it to users from day zero.

Pricing and deployment details

Together AI said Muse Spark Open is available via its Together API and playground. The company listed pricing at $1.25 per 1 million input tokens, $4.25 per 1 million output tokens and $0.15 per 1 million cached input tokens. Meta’s broader release notes also indicate deployment options include self-hosting and hosted APIs, and that inference support extends across a range of partners including Together AI.

The significance of the launch is that one of Meta’s newest open-weight models is now available through a third-party inference provider on release day, alongside Meta’s own distribution channels. The evidence provided does not include usage data, customer adoption figures or independently verified benchmark results, so claims about competitive performance remain those of the companies involved.