Meta’s Muse AI models arrive on Google Cloud Model Garden

Meta's Muse AI models arrive on Google Cloud Model Garden

Meta’s Muse Ember, Muse Spark 1.3 and Muse Wonder are now available through Google Cloud’s Model Garden via managed APIs, and Sundar Pichai publicly welcomed them on X. The arrangement is distribution, not a merger or joint model-development deal, and Meta’s reported comparison puts Muse Ember at about $0.14 per task versus $0.41 for Claude Haiku 5.5.

What Happened

Google Cloud customers can now reach three Meta AI models through Google’s own platform, after Meta moved its Muse family beyond its own ecosystem. Mark Zuckerberg announced the availability on X, and Google CEO Sundar Pichai replied publicly.

The models are Muse Ember, Muse Spark 1.3 and Muse Wonder. All three sit in Google Cloud’s Model Garden, the catalogue that gives businesses and developers access to AI models from different providers.

Managed application programming interfaces let companies connect those models to their own apps without running the underlying model infrastructure themselves.

Pichai welcomed the arrival directly on X, writing: “Congrats and welcome to the model garden, excited to serve our Google Cloud customers!” According to thedigitalbriefnews.com, his reply congratulated Meta and confirmed the business relationship behind the launch.

Muse Wonder is positioned as the most capable model of the group. The Times of India reports it offers native computer use and long-context understanding across browsers, desktop and mobile applications.

It also features sub-agent delegation, parallel tool calling and agentic memory, and is aimed at multi-file repository updates and end-to-end workflow automation.

Muse Spark 1.3 is the workhorse agentic model for long-horizon workflows, balancing intelligence, speed and scalable performance. The Times of India reports it supports native tool calling and Model Context Protocol server integration, which lets developers build agents that complete multi-step tasks.

Muse Ember handles the repeated, everyday work that agents generate all day, such as tool execution, routing, document analysis and everyday coding subtasks.

The performance claims are Meta’s own. Gigazine reports Muse Ember scores 36 points on Artificial Analysis’ Intelligence Index, matching Claude Haiku 5.5 at its highest inference setting and ahead of OpenAI’s GPT-6 Luna at 28 points.

Meta’s published comparison puts Ember’s cost per task at about $0.14, roughly one-third of Claude Haiku 5.5’s $0.41.

Benchmark detail varies by task type. Gigazine reports Ember scored 37.8% on AutomationBench-AA against Claude Haiku 5.5’s 13.9% and GPT-6 Luna’s 20.2%, and 40.9% on Terminal-Bench 4.0 against 39.2%. It trailed on FrontierCode 2.0-AA, at 27.2% versus Claude Haiku 5.5’s 34.9%.

Meta’s figures show Spark scoring 37 on the same index against Ember’s 36, at about $2.39 per task versus $0.14.

thedigitalbriefnews.com notes that the Claude Haiku comparison is a reported one, not an independent benchmark result. The outlet also states the exchange does not indicate a merger, joint AI-model development agreement or change in ownership.

The Business Today report covering the announcement was updated on November 12, 2026.

What This Means For You

For developers and engineers: Because three Meta models now sit behind Google Cloud’s managed APIs, engineers could test agent and coding workflows without provisioning model infrastructure themselves.

For business leaders and founders: Because Model Garden mixes Meta’s models with Google’s own Gemini line on one platform, cloud purchasing decisions could hinge on catalogue breadth rather than a single vendor’s family.

For students and researchers: Because Ember accepts a 1 million-token context window and up to 512,000 output tokens according to Gigazine, long codebase or document analysis could run on a cheaper tier.

Because Muse Wonder supports native computer use, multi-step research tasks across apps become a plausible test case.

Why It Matters

This suggests cloud catalogues are becoming the shopfront for model choice, not just raw compute. Google continues to sell its own Gemini models while hosting models built by a direct competitor in the same market.

For buyers, the practical question moves from “which model” to “which catalogue plus which tier”.

Meta’s willingness to distribute through a rival’s cloud also widens reach beyond its own apps. That could mean more developers encounter Muse models by default, without a separate vendor relationship.

Key Takeaway

  • Google Cloud’s Model Garden now lists Meta’s Muse models, giving developers a managed route to a rival’s AI family.
  • Pichai’s public reply signalled Google’s willingness to host models built by a competitor.
  • Meta’s reported cost figures for Ember may prompt teams to recheck agent-heavy budgets.
  • thedigitalbriefnews.com states the arrangement is distribution, not a merger or joint model-development deal.

Which Meta models are available on Google Cloud?

Three from Meta’s Muse family, each aimed at different workloads, from high-volume routine tasks to complex software engineering.

Is this a partnership between Google and Meta on AI development?

The arrangement is distribution. thedigitalbriefnews.com states the exchange does not indicate a merger, joint AI-model development agreement or change in ownership, with Meta continuing to develop its Muse models.

How does Muse Ember’s cost compare with rival models?

Gigazine reports Meta’s comparison puts Muse Ember at roughly $0.14 per task against about $0.41 for Claude Haiku 5.5. thedigitalbriefnews.com notes that comparison is reported, not an independent benchmark result.

Sources

This report draws on coverage from Business Today, thedigitalbriefnews.com and gigazine.net.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

Click on below button to add AICopse for your Preferred Source

Add as a preferred source on Google







Join Our Newsletter

Get articles and updates delivered straight to your inbox regularly.

No spam ever. Unsubscribe anytime easily.