Meta’s Muse Ember, Muse Spark 1.3, and Muse Wonder land on Google Cloud

Google Cloud now offers three of Meta’s Muse models: Muse Ember, Muse Spark 1.3, and Muse Wonder. Customers can reach them through Google’s managed APIs on the Gemini Enterprise Agent Platform.

Three models, three jobs

Muse Ember is the volume model. Meta announced it on November 9, 2026, and describes it as its smallest and most efficient model to date.

Ember is aimed at high-volume agentic execution. Agentic work means an AI that takes actions, such as calling tools or running multi-step tasks, rather than only answering questions.

According to Artificial Analysis benchmarks, Ember reportedly performs on par with models like Anthropic’s Claude Haiku 5.5. It reportedly does so at one-third the cost per task.

Muse Spark 1.3 is the coding-focused update. It arrived on September 2, 2026, and builds on the earlier Spark release.

Compared with its predecessor, Spark 1.3 cuts tool calls by approximately 20% and token usage by approximately 25%.

Muse Wonder sits at the top of the stack. Meta unveiled it on October 12, 2026, and built it for complex reasoning tasks.

What they share

All three models support a 1 million token context window and accept multimodal inputs. That covers text, images, video, audio, and PDFs.

Google is not the only cloud carrying the lineup. The models are also available on Oracle Cloud.

From Llama to Muse

The Muse family marks a shift from Meta’s earlier Llama line of models. Muse moves in a different direction: offering models through managed APIs means customers call them as a hosted service rather than downloading and running them themselves.

The release cadence has been brisk. Spark 1.3 shipped in September, Wonder followed in October, and Ember arrived in November.

What this means for the AI market

Google builds Gemini, and Gemini competes directly with Muse. Yet Google is now selling Meta’s models to its own cloud customers through a platform carrying the Gemini name. Cloud providers make money when customers run workloads on their infrastructure, regardless of whose model powers the work.

For Meta, the gain is reach. Enterprises often prefer to buy through cloud vendors they already have contracts with. Appearing on Google Cloud and Oracle Cloud puts Muse in front of buyers who might never sign a separate deal with Meta.

Ember targets the cheap, fast tier where models like Claude Haiku 5.5 operate. Wonder targets the premium reasoning tier. Meta is contesting both ends of the market at once.