Meta's Muse Models Arrive on Google Cloud as Pichai Welcomes a Gemini Rival

Google chief executive Sundar Pichai welcomed the move publicly. "Congrats and welcome to the model garden, excited to serve our Google Cloud customers!" he wrote, according to The Hans India.

The practical change is in how companies buy the models. Firms that already run workloads and spending commitments on Google Cloud can now call Meta's models without signing a separate contract with Meta. On Monday OpenAI made a similar move, listing ChatGPT Enterprise on AWS Marketplace so buyers can count eligible spending toward their Amazon commitments. Google earns cloud revenue whichever model a customer picks, and Meta reaches enterprise buyers who prefer to stay with their existing cloud supplier.

The listing covers three models. Muse Wonder, released on October 12, is Meta's frontier model, built for demanding software engineering and long-running agent work. Muse Spark 1.3, an update released on September 2, is aimed at long agentic workflows and has native tool calling and support for MCP servers. Muse Ember, which Meta released on November 9, is its smallest and cheapest model, built for repetitive, high-volume agent tasks. All three accept multimodal input and support context windows of up to one million tokens.

Meta has said Ember performs comparably to Anthropic's Claude Haiku 5.5 at about a third of the cost per task. Those are Meta's own figures. OrcaRouter, a service that routes requests between AI models, said Meta's benchmark results for Ember had not been independently reproduced.

The Muse line moves Meta away from its Llama family, which developers could download and run themselves. Muse is offered mainly as a hosted service that customers pay to call. The models are also available on Oracle Cloud, according to Crypto Briefing.

OrcaRouter's review of Google's documentation found that the Google Cloud listing is a preview. It is served from a single global endpoint, with no regional endpoints listed for any of the three models. No service-level agreement or capacity guarantee has been published, and Google's documentation describes the models as "not a Google product", so customers must take output problems to Meta.

Google has not published its own rate card for the models. According to OrcaRouter, which cited Meta's list prices, Muse Wonder costs $4.00 per million input tokens and $20.00 per million output tokens. Muse Spark 1.3 costs $1.25 and $4.25, and Muse Ember costs $0.10 and $0.50.

Meta and Google have not said when the models will leave preview or whether regional hosting will follow. For companies that must keep data in a particular country, that will decide whether the listing is usable.