Meta's Muse models land on Google Cloud, inside a Gemini-branded storefront

Google Cloud customers can now call Muse Ember, Muse Spark 1.3 and Muse Wonder through Google's managed APIs on the Gemini Enterprise Agent Platform, according to Crypto Briefing. Oracle Cloud carries the same lineup, so nobody has to sign a standalone deal with Meta to get at them. It's a rival's AI sitting on Google's shelf, under a platform wearing the Gemini name.

Ember is the cheap, fast one. Meta dropped it on November 9, 2026, billing it as the smallest and most efficient model it has shipped, built for high-volume agentic execution — the kind of work where an AI calls tools and runs multi-step tasks instead of just answering questions. Artificial Analysis benchmarks have it performing on par with Anthropic's Claude Haiku 5.5 at roughly one-third the cost per task.

Spark 1.3 is the coding play, launched September 2, 2026 as an update to the original Spark. Against its predecessor it trims tool calls by about 20% and token usage by about 25%. Wonder, out October 12, 2026, takes the top of the stack for complex reasoning. All three swallow a 1 million token context window and handle text, images, video, audio and PDFs.

The Muse family is a deliberate break from Llama. Customers call these models as a hosted service through managed APIs rather than downloading and running them on their own hardware. And Meta didn't ease in: Spark 1.3 in September, Wonder in October, Ember in November.

The math works even for the company being undercut. Cloud providers collect when workloads run on their infrastructure, whoever's model is doing the thinking — and enterprises tend to buy through vendors they already have contracts with. Ember goes after the budget tier where Claude Haiku 5.5 lives, Wonder goes after premium reasoning. Meta is swinging at both ends at once.