Meta models on Google Cloud
Deploy Meta’s latest models via the model garden on Google Cloud to build production-ready AI agents and applications. With a range of model sizes and capabilities, you can choose the right model for your specific use case—from lightweight, efficient models to multimodal versions. Access these models as serverless APIs and use their native multimodal and multilingual abilities for highly efficient text and visual intelligence.
Meta’s expanded model portfolio provides developers with the right intelligence for every step—spanning state-of-the-art proprietary models like Muse Wonder model series and Muse Spark model series for complex engineering and multi-step agentic workflows, alongside open-weight models like Muse Spark Open and Llama models that offer the transparency and flexibility needed for custom deployments. When you build with Meta models in the model garden on Google Cloud, you can seamlessly discover, customize, and deploy both cutting-edge proprietary and open-weight models, combining Meta's frontier capabilities with Google Cloud's enterprise-grade security, cost-efficient scalability, and comprehensive managed AI tools.
Frontier agentic models, built for complex engineering
Muse Wonder is Meta’s most capable frontier model, purpose-built for complex software engineering, offering an optimal balance of speed, reliability, and deep persistence. Muse Wonder offers native computer use and long-context understanding, enabling the execution of sophisticated, multi-step workflows across browsers, desktop, and mobile applications. Muse Wonder models offer state-of-the-art coding, orchestration, and multimodal reasoning capabilities, and feature sub-agent delegation, parallel tool calling, and agentic memory.
Top use cases include real-world software engineering, multi-file repository updates, and end-to-end workflow automation. It can also operate everyday applications, support complex legal and professional knowledge work, and reason over vast codebases.
Start building
Workhorse agentic models, built for action at scale
Muse Spark 1.3, available on Google Cloud, is a workhorse agentic model designed for long-horizon workflows, offering an optimal balance of intelligence, speed, and scalable performance. Muse Spark 1.3 offers native tool calling and MCP server integration, enabling the creation of sophisticated AI agents that complete multi-step tasks across professional and personal workflows. Muse Spark models offer coding, reasoning, and autonomous execution capabilities, and feature in-series specialization optimized for high-volume production environments on Gemini platform.
Top use cases include multi-step workflow automation, tool-using application development, and long-horizon agentic execution. It can also run personal and enterprise assistants, orchestrate complex API and MCP pipelines, accelerate software development, and reason over dynamic, real-world tasks at scale.
Start building
Lightweight execution models, built for high-volume workflows
Muse Ember on Google Cloud is an efficient, multimodal execution model built for repeated everyday work, offering an optimal balance of capability, throughput, and cost-efficiency. Muse Ember offers comprehensive text, image, audio, and video processing, enabling developers to match task complexity to model capability across enterprise pipelines. Muse Ember models offer coding, tool-calling, and document-analysis capabilities, and feature a lightweight architecture purpose-built to complement Muse Wonder and Muse Spark on Google Cloud.
Top use cases include everyday coding subtasks, automated tool execution, and large-scale document understanding. It can also power high-volume task routing, serve as the execution layer for complex multi-agent orchestration, analyze multimodal media, and keep long-running workflows moving at scale.
Start building
Muse Spark Open: Natively multimodal open weights for agentic work
Muse Spark Open, deployable on Google Cloud, is an 816B-parameter (42B active) MoE open-weight model with a 1M-token context window, purpose-built for long-horizon coding and agentic workflows. Accepting text, image, video, and audio inputs across 100+ languages, it excels at multi-step planning, resilient tool invocation, real-world software engineering, synthetic data generation, and LLM-as-a-judge evaluation on the Gemini platform.
Developers on Google Cloud can also access the Llama 4 and Llama 3 model families for general-purpose multimodal and text generation. Optimized for speed and cost-efficiency, Llama models support foundational use cases including language translation, content creation, visual search, and customer service chatbots.
Start building
Muse Wonder
Frontier agentic models, built for complex engineering
Muse Wonder is Meta’s most capable frontier model, purpose-built for complex software engineering, offering an optimal balance of speed, reliability, and deep persistence. Muse Wonder offers native computer use and long-context understanding, enabling the execution of sophisticated, multi-step workflows across browsers, desktop, and mobile applications. Muse Wonder models offer state-of-the-art coding, orchestration, and multimodal reasoning capabilities, and feature sub-agent delegation, parallel tool calling, and agentic memory.
Top use cases include real-world software engineering, multi-file repository updates, and end-to-end workflow automation. It can also operate everyday applications, support complex legal and professional knowledge work, and reason over vast codebases.
Start building
Muse Spark 1.3
Workhorse agentic models, built for action at scale
Muse Spark 1.3, available on Google Cloud, is a workhorse agentic model designed for long-horizon workflows, offering an optimal balance of intelligence, speed, and scalable performance. Muse Spark 1.3 offers native tool calling and MCP server integration, enabling the creation of sophisticated AI agents that complete multi-step tasks across professional and personal workflows. Muse Spark models offer coding, reasoning, and autonomous execution capabilities, and feature in-series specialization optimized for high-volume production environments on Gemini platform.
Top use cases include multi-step workflow automation, tool-using application development, and long-horizon agentic execution. It can also run personal and enterprise assistants, orchestrate complex API and MCP pipelines, accelerate software development, and reason over dynamic, real-world tasks at scale.
Start building
Muse Ember
Lightweight execution models, built for high-volume workflows
Muse Ember on Google Cloud is an efficient, multimodal execution model built for repeated everyday work, offering an optimal balance of capability, throughput, and cost-efficiency. Muse Ember offers comprehensive text, image, audio, and video processing, enabling developers to match task complexity to model capability across enterprise pipelines. Muse Ember models offer coding, tool-calling, and document-analysis capabilities, and feature a lightweight architecture purpose-built to complement Muse Wonder and Muse Spark on Google Cloud.
Top use cases include everyday coding subtasks, automated tool execution, and large-scale document understanding. It can also power high-volume task routing, serve as the execution layer for complex multi-agent orchestration, analyze multimodal media, and keep long-running workflows moving at scale.
Start building
Open-weight models
Muse Spark Open: Natively multimodal open weights for agentic work
Muse Spark Open, deployable on Google Cloud, is an 816B-parameter (42B active) MoE open-weight model with a 1M-token context window, purpose-built for long-horizon coding and agentic workflows. Accepting text, image, video, and audio inputs across 100+ languages, it excels at multi-step planning, resilient tool invocation, real-world software engineering, synthetic data generation, and LLM-as-a-judge evaluation on the Gemini platform.
Developers on Google Cloud can also access the Llama 4 and Llama 3 model families for general-purpose multimodal and text generation. Optimized for speed and cost-efficiency, Llama models support foundational use cases including language translation, content creation, visual search, and customer service chatbots.
Start building
Benefits and capabilities of Meta models on Google Cloud
Accelerate AI development
Gemini platform provides an integrated environment to evaluate, deploy, and manage Meta model-enabled applications quickly and at scale.
Optimize performance and cost
Simplify how you deploy and scale Meta models with a fully managed infrastructure designed for AI workloads, and the option to select from flexible pricing models like dedicated endpoints, or pay-as-you-go pricing.
Build sophisticated AI agents
Develop agents with Agent Development Kit and the advanced capabilities of Meta models.
Built-in security, compliance, and data governance
Leverage Google Cloud's built-in security, privacy, data governance, and compliance capabilities tailored to adhere to enterprise-level standards.
Maximize the power of your data
Integrate your enterprise data with Meta's models' advanced capabilities, leveraging tools like BigQuery to extract valuable insights and drive informed decision-making.
Enhanced capabilities
Meta models demonstrate advanced abilities in complex reasoning, vision analysis, code generation, and multilingual processing. These models can follow intricate instructions and generate nuanced, comprehensive outputs.
