Muse Wonder

• 🎯 Meta's frontier reasoning model for agentic work
• 📏 Massive 1M-token context window
• 👁️ Accepts text, images, video, audio, and PDFs
• 🔧 Parallel tool calling, sub-agent delegation, error recovery
• 💻 Code-optimized with built-in web search
• 🧠 Long-horizon tasks run without step-by-step direction

Meta

Meta Platforms is an American multinational technology company headquartered in Menlo Park, California. Best known for operating Facebook, Instagram, WhatsApp, and Threads, Meta has become one of the most influential players in open-weight AI research through…

Muse Wonder is Meta's frontier reasoning model, built for long-horizon agentic work rather than single-turn chat. Released in October 2026, it accepts an unusually broad set of inputs — text, images, video, audio, and PDF documents — and returns text, all within a 1,048,576-token context window that makes whole-repository, whole-casefile, or whole-archive reasoning practical in a single session.

Within Meta's lineup it occupies the frontier reasoning tier, sitting well above the open-weight Llama text models such as Llama 3.3 70B and the compact Llama 3.2 3B, and complementing the image-generation side of the Muse family in Muse Image. Its distinguishing behaviour is agentic autonomy: it plans multi-step work, delegates to sub-agents, issues tool calls in parallel, searches the web, and recovers from failed steps instead of stalling.

That combination makes Muse Wonder best suited to complex, multi-hour tasks where a user states a goal rather than a sequence of instructions — research synthesis across large document sets, multimodal analysis, software engineering spanning many files, and tool-driven workflows that need to run to completion unattended.