Meta's Muse Wonder lands on AI Gateway: one key, 1M-token context for coding and long-running agents

Meta's Muse Wonder model is now available through AI Gateway, aimed at coding, long-running agents, and computer use. Reaching the model previously meant creating a separate Meta account and its credentials; access now runs through one AI Gateway API key. The AI Gateway page lists the model in its catalog and shows calls using meta/muse-wonder-1.0 as the model name.

The model can plan multi-step tasks, delegate work to sub-agents, and recover from errors as work progresses. Those behaviors target long-running agent workflows rather than single-turn question answering. The model description names coding, long-running agents, and computer use as the design uses, and lists multi-step planning, sub-agent delegation, and error recovery.

The model offers a 1M-token context window and supports text, image, and PDF input, tool calling, and structured output. Agents can therefore work with code, screenshots, and documents inside one workflow instead of switching pipelines per modality. The model description lists the context length, the input modalities, and support for tool calling and structured output together.

Access is set up by placing AI_GATEWAY_API_KEY in the environment and calling the model as meta/muse-wonder-1.0. Coding agents can instead run npx vercel ai-gateway setup and select the model in the agent's model settings. The page provides a streamText call example and command-line setup steps for coding agents.

This report is for developers and engineers choosing a model for coding or long-running agent workflows, and it covers Muse Wonder's availability on AI Gateway, how it is called, and the capabilities listed. The model description scopes the uses to coding, long-running agents, and computer use, and lists a 1M-token context window, text, image, and PDF input, tool calling, and structured output. The access path is to set AI_GATEWAY_API_KEY and call meta/muse-wonder-1.0, while coding agents can run npx vercel ai-gateway setup and select the model in their model settings.

The model description gives no benchmark results, latency, or cost figures, and does not say on which tasks or at what scale multi-step planning, sub-agent delegation, and error recovery were validated. The listed capabilities appear as design uses without a controlled comparison. This report rests on the AI Gateway model description and call examples, with no independent reproduction.