Muse Wonder
Meta’s first frontier-class model, built for sustained agentic work.
Muse Wonder is Meta Superintelligence Labs’ most capable API model, released October 12, 2026 for long, complex personal and professional agent tasks. It takes text, images, audio, video and PDFs. Meta highlights Active Memory, which carries information across conversations, and Active Focus, which recovers from errors in multi-step tasks. It is available through the Meta Model API and Muse Code.
- Provider
- Meta
- Type
- Language model
- Released
- Oct 12, 2026
- Input
- text, image, file, audio, video
- Output
- text
- Open weights
- No
Best for
- Agents
- Coding
- Reasoning
- Multimodal
- Long context
Strengths
- Active Memory and Active Focus
- Audio and video input
More from Meta
- Llama 4 Maverick — Meta. Meta’s 128-expert open multimodal MoE model.
- Llama 4 Scout — Meta. Efficient open Llama 4 model with a very long native context.
- Muse Spark Open — Meta. Meta’s open-weight 816B MoE, tuned for agentic coding.
- Muse Spark 1.3 — Meta. Meta’s multimodal reasoning model for long-running agents and coding.
- Muse Image — Meta. Meta’s text-to-image model from the Muse family.
- V-JEPA 2 — Meta. Meta’s open JEPA world model for video understanding, prediction and robot planning.
- Llama Guard 4 12B — Meta. Open multimodal safety classifier for prompts and responses.
Tools that use it
- Muse Code — Meta. Meta’s terminal coding agent, priced to undercut rivals.