Meta's first frontier-class AI model — a 1M-token context window, an Active Memory system that carries context across conversations, and an Active Focus mode for grinding through hard, multi-step agentic tasks. Ships as Muse Wonder (API + Muse Code) and the larger Muse Wonder Full (Muse Code only).
Last verified · 2026-10-12 · by Moe Ameen
Muse Wonder is Meta Superintelligence Labs' first frontier-class model, launched October 12, 2026 alongside a larger sibling, Muse Wonder Full. It builds on the Muse Spark family and is pitched for long-running, multi-step agentic work rather than one-shot chat. Meta calls the pair its first frontier-class models, says both were trained at its gigawatt-scale Prometheus data center, and describes Muse Wonder as its most aligned model to date, with its strongest safeguards yet against CBRN and cyber misuse.
The two defining capabilities are built for sustained work. Active Memory retains relevant context across conversations and brings it back when a task needs it; Active Focus keeps the model pushing toward a hard goal — trying alternative approaches, recovering from errors, and checking in before important or irreversible actions. Both models carry a 1 million-token context window. Meta cites domain expertise in law, finance, and health, frontier coding, and the ability to work in a browser and across apps. On modality, the models are text-first with image inputs; Meta's demos build artifacts from photos, like designing custom merchandise from images in an Instagram profile or generating a 3D game from image inputs.
Access splits by model. Muse Wonder is available on the Meta Model API and in Muse Code (Meta's coding agent), with reported pricing of $4 per million input tokens and $20 per million output — a step up from the Muse Spark line's $1.25/$4.25. Muse Wonder Full is exclusive to Muse Code subscribers and is not offered via API. On Artificial Analysis's Intelligence Index, Muse Wonder Full scores 56 — tying Claude Sonnet 5.5 and topping GPT-6 Astra, Gemini 4 Argon, and Claude Fable 5.1 (53 each) — while the standard Muse Wonder scores 52, trailing GPT-6 Astra on that same measure. Only Claude Opus 5.5 (58) scores higher than Wonder Full, and Meta frames its biggest gains as concentrated on agentic and coding benchmarks rather than broad world knowledge.
The honest framing: Muse Wonder is a reasoning-and-agent model, not a content-production tool. It can draft copy, write a script, reason over a long document, or describe an image, but it hands back raw text or plans in an agent session or an API response. It does not keep a brand voice across a week of posts, render a carousel or an avatar video, size anything per platform, or schedule and publish to your social channels. Treat the specific benchmarks and pricing as a launch-window snapshot — verify current numbers on Meta's own pages.
The most creator-relevant thing about Muse Wonder is the combination of a 1 million-token window, Active Memory, and image inputs: you can hand it a full quarter of source material — a long webinar transcript, a batch of client notes, a research PDF, and a folder of reference screenshots — and it will hold all of it at once and pull out the structure. The key moments, the quotable lines, the angles worth a post, the section outline. That is genuinely strong raw material, and it is a better research pass than the Muse Spark line could do. But it stops at a plan written in text. It will tell you "the strongest segment is the 12-minute mark and here are the five hooks," but it will not cut that segment into a vertical clip, reframe it to keep the speaker centered, burn in animated captions, or push it to TikTok. That span — between a smart plan and finished, shipped media — is exactly what Kompozy covers.
Here is the concrete handoff, and it is deliberately different from wiring up the pricey Meta Model API yourself. Take the outline and hook list Muse Wonder produced from your long source and bring it into Kompozy as a content brief. Kompozy's Clipped Shorts turn the actual long-form video into vertical cuts; the same brief fans into a brand-exact Carousel, Quote Graphics built from the lines the model surfaced, Persona Shorts with a face-locked avatar, a Blog Article, an Email Newsletter, and native Text Posts — all held to one voice by your Persona Brief and banned-word filters, and each rendered as finished media, not a description of it. Then Kompozy schedules and publishes the whole set across eight social platforms plus blog and email from one queue with Autopilot and a per-post review pipeline. And because Kompozy's own generation runs on managed Claude and OpenAI models — the same frontier class Muse Wonder competes in — you can skip the research hop entirely and brief Kompozy directly. Muse Wonder reads your material and tells you what to make; Kompozy is what turns that into the on-brand posts and gets them live.
Muse Wonder is Meta Superintelligence Labs' first frontier-class AI model, launched October 12, 2026 alongside the larger Muse Wonder Full. Both carry a 1 million-token context window and are built for agentic work, with an Active Memory system that carries context across conversations and an Active Focus mode for sticking with hard, multi-step tasks. It is available on the Meta Model API and in Muse Code.
Muse Wonder is on the Meta Model API and in Muse Code at a reported $4 per million input tokens and $20 per million output — a step up from the Muse Spark line. The larger Muse Wonder Full is exclusive to Muse Code subscribers and is not offered via API. Treat the pricing as a launch-window figure and confirm current numbers on Meta's page.
No. Muse Wonder is text-first with image inputs — it reasons over images you feed it and builds artifacts from them, but it outputs text and plans, not generated images, video, or audio. To turn what it drafts into finished, published media, pair it with a content engine like Kompozy that renders the media and publishes across platforms.
Muse Wonder is a frontier-class jump, not an iteration. On Artificial Analysis's Intelligence Index it scores 52 (standard) and 56 (Full), plus it adds the new Active Memory and Active Focus systems that Muse Spark 1.3 doesn't have. It is also pricier — a reported $4/$20 per million tokens against Muse Spark's $1.25/$4.25.