Meta Muse Wonder - Meta's first frontier AI model explained

Meta has finally joined the frontier race. On October 12, 2026, Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full — the company's first models it calls frontier-class. But there's a twist: there are two models, and the better one isn't available on the API. Here's everything you need to know.

What Is Muse Wonder?

Muse Wonder is a large language model developed by Meta Superintelligence Labs (led by Alexandr Wang), released October 12, 2026, and positioned by Meta as "a frontier model for personal and professional AI agents." It succeeds the lightweight Muse Spark family (Muse Spark 1.3 scored 48 on the Artificial Analysis Intelligence Index six weeks earlier) and goes the opposite direction: bigger, more capable, and priced squarely against the top tier from OpenAI and Anthropic.

Meta built both models at its gigawatt-scale Prometheus data center in Ohio, with a larger base model and additional reinforcement learning compared to the Spark line. The model card notes it generalizes better than expected to tasks outside its post-training targets.

Like every frontier model this fall, Wonder is multimodal on input: it accepts text, images, video, and audio (PDFs via file upload) and generates text output. It supports function calling, JSON-schema outputs, built-in web search, and six effort settings ranging from minimal up to max — so developers can trade cost against capability per call.

For context, if you're tracking the crowded frontier leaderboard, our ChatGPT vs Claude vs Gemini comparison breaks down how the other contenders stack up.

Two Models, Two Doors: Muse Wonder vs Muse Wonder Full

The single most confusing thing about this launch is the naming. There are two models, and access depends on where you buy:

  • Muse Wonder (model ID muse-wonder-1.0) — available on the Meta Model API for developers and inside Muse Code, Meta's subscription coding and agent product.
  • Muse Wonder Full — the larger, more capable variant. It is not sold through the API at all. You can only get it inside Muse Code, on the two highest-usage plans: Power Usage ($50/month) or Maximum Usage ($200/month).

This matters: cranking effort settings on the API's Wonder to "max" does not give you Wonder Full. Meta treats Full as a separate, more capable model. If your work needs the strongest version and you're on the pay-as-you-go API, you simply can't have it — you'd need a Muse Code subscription.

Key Features: Active Memory and Active Focus

Meta's pitch centers on two named features that build on its earlier open-source work on long-term AI memory (see our coverage of NVIDIA's local agent models for a competing approach):

Active Memory

Active Memory lets the model carry what matters to you across conversations — key facts, preferences, and project context persist without you re-explaining them every session. Meta claims it leads its own internal MemoryBench, its in-house memory benchmark.

Active Focus

Active Focus is meant to keep an agent working on a hard task: sustaining effort on long-running work, retrying different approaches when stuck, and checking in before doing anything important. If you're building agents that run for minutes or hours, this is the feature aimed squarely at you.

Both models ship with a context window of roughly one million tokens (1,048,576 on the API version of Wonder), which gives agents and coding workflows the room they need to work with entire codebases or long document sets.

Pricing: How Much Does Muse Wonder Cost?

Through the Meta Model API, Muse Wonder costs:

  • $4 per million input tokens
  • $0.20 per million cached input tokens
  • $20 per million output tokens

That's the exact headline rate as Claude Opus 5.5 — Meta is deliberately matching Anthropic's flagship price. For comparison, our Claude Haiku 5.5 guide shows the other end of the market: $0.10 per million input tokens for Anthropic's cheapest model.

Muse Wonder Full has no API price because it's not on the API — it comes bundled in Muse Code's Power Usage ($50/month) and Maximum Usage ($200/month) plans.

One independent cost test (cited by myclaw.ai) found Wonder coming in cheaper per completed task than Anthropic's top models in one scenario — worth noting, but it's a single test, not a trend.

Benchmarks: How Good Is It Really?

Meta's own testing, laid out in its model card, puts Wonder and Wonder Full in the same range as Claude Fable 5.1 and GPT-6 Astra on coding, legal, and financial tasks. Treat that with care — the numbers come from Meta, and a full safety report is "coming soon," according to the company.

Independent testing tells a more measured story. Artificial Analysis, an independent benchmarker, scores Muse Wonder at 52 and Wonder Full at 56 on its Intelligence Index — versus 58 for Claude Opus 5.5. So Meta's best lands second in the world, two points behind Anthropic's flagship and ahead of GPT-6 Astra (53) and Gemini 4 Argon. That's genuinely impressive for Meta's first attempt — but it isn't the lead.

Also worth remembering: Wonder is in preview at launch. Expect rough edges, possible regional availability limits, and changing behavior as Meta iterates.

Who Is Muse Wonder For?

  • Developers building long-running agents: the 1M-token context, six effort levels, and Active Focus make it a serious option for agentic workflows.
  • Coding workflows: Muse Code subscribers get both models, including the stronger Full variant that API users can't touch.
  • Teams already in the Meta stack: if your data and tooling live on Meta infrastructure, a frontier model from the same vendor simplifies procurement.
  • Domain-heavy work: Meta positions both models for commercial and research use in law, finance, and health knowledge work.

If you just want a chatbot to talk to, Wonder isn't aimed at you — for now it's developer- and agent-focused, not a replacement for the Meta AI app most people use.

Pros and Cons

Pros

  • Meta's first genuinely frontier-class model — second place on the independent index is a real achievement.
  • Active Memory and Active Focus directly address the two biggest pain points of agent builders.
  • 1M-token context and six effort levels give developers fine-grained cost/capability control.
  • Multimodal inputs (text, images, video, audio) with function calling and web search built in.

Cons

  • The better model (Wonder Full) is locked behind a $50+/month subscription — API users get the weaker variant.
  • Still trails Claude Opus 5.5 on independent benchmarks; no independent testing of Meta's memory claims yet.
  • Preview status: expect instability, and the full safety report hasn't landed.
  • Closed weights — after two years of insisting on open models, Meta's frontier play is paywalled, priced like GPT-4-class access rather than anything Llama ever charged.

Verdict: Should You Try It?

If you build long-running agents or coding workflows and want a 1M-token-context model priced at flagship rates, Muse Wonder is worth a serious look — independent cost tests even suggest it can beat Anthropic's best per task. But wait if you need a stable, non-preview model, proven regional access, or the strongest variant without a $50/month commitment.

Wonder is Meta's declaration that the open-source-first era is ending: this is the company finally re-entering the race it watched from the sidelines. Whether it deserves your workload will come down to the independent benchmarks over the coming months.

And if you have a strong opinion about which AI deserves the crown, you can vote in our AI Showdown poll — voting is open through the end of the year.

FAQ

Is Muse Wonder the same as Muse Spark?

No. Muse Spark was Meta's lightweight, low-latency family; Wonder is the new bigger, frontier-class line built for complex coding and agent tasks.

Can I use Muse Wonder Full on the API?

No. Muse Wonder Full is only available to Muse Code subscribers on the Power Usage ($50/month) or Maximum Usage ($200/month) plans. The API sells Muse Wonder only.

How much does Muse Wonder cost?

$4 per million input tokens, $0.20 per million cached input tokens, and $20 per million output tokens on the Meta Model API — matching Claude Opus 5.5's headline rate.

How does Muse Wonder compare to Claude Opus 5.5?

On the Artificial Analysis Intelligence Index, Muse Wonder scores 52 and Wonder Full 56, versus 58 for Claude Opus 5.5 — so Meta trails by a narrow margin, while matching Opus on price.