Muse Wonder Explained: Meta’s Frontier Model, Pricing & Benchmarks

Disclosure: The Unfiltered AI is reader-supported. This article contains no affiliate links — it’s a straight explainer based on Meta’s published documentation and independent benchmark reporting, verified October 16, 2026.

The short version

Muse Wonder is Meta’s first real frontier AI model, released October 12, 2026 by Meta Superintelligence Labs. It costs $4 per million input tokens and $20 per million output tokens — exactly what Anthropic charges for Claude Opus 5.5. A stronger version, Muse Wonder Full, exists but is only available through a Muse Code subscription.

What is Muse Wonder?

Muse Wonder is a family of two large reasoning models built for long-horizon agent work and coding. Both were trained at Meta’s gigawatt-scale Prometheus data center and support a ~1-million-token context window with text, image, video, audio, and PDF input.

Meta’s pitch centers on two capabilities: Active Memory (carrying context about you across conversations for weeks or months) and Active Focus (sticking with hard problems, trying different approaches, asking clarifying questions before acting).

Notably, unlike Meta’s earlier Llama models, Muse Wonder is closed — no downloadable weights, API and Muse Code only. The open-source era of Meta AI pricing is over.

Muse Wonder pricing (verified Oct 16, 2026)

Muse Wonder Muse Wonder Full
Input / 1M tokens $4.00 Subscription only (Muse Code Power $50 / Maximum $200 plans)
Output / 1M tokens $20.00
Cached input / 1M $0.20
API model ID muse-wonder-1.0 Not on the API
Context window ~1,048,576 tokens

For context, the previous Muse Spark 1.3 cost $1.25/$4.25 — Wonder is roughly 3x the price per task on independent testing.

How does it benchmark?

On Artificial Analysis’ independent Intelligence Index: Muse Wonder Full scores 56, plain Wonder 52 — behind Claude Opus 5.5 (58), ahead of GPT-6 Astra (53). Meta’s own 14-benchmark table is friendlier (Wonder family leads 7 rows, Opus 5.5 leads 6), but vendor charts always are. The fair summary: a genuine frontier model, close to the leaders, not a clear winner. Strengths cluster in long-running agent work (SWE-Marathon, DeepSWE); factual knowledge tests are a relative weak spot.

How to get access

  • Muse Wonder: Meta Model API (muse-wonder-1.0), also via OpenRouter and every Muse Code plan
  • Muse Wonder Full: Muse Code Power ($50) and Maximum ($200) plans only — no public API
  • Both will power Meta’s Muse personal agent, rolling out to subscribers first

Should you switch?

Test before you switch. Run Wonder on 30–50 of your own real tasks next to your current model and compare success rate, cost per finished task, and speed. At $4/$20 it’s priced identically to Opus 5.5, so the decision comes down to which model wins on your workloads — particularly long-horizon agent tasks, where Wonder looks strongest. Keep a fallback ready either way.

FAQ

Is Muse Wonder open source?

No. Unlike Llama, Meta has not released the weights and hasn’t said whether it ever will.

What’s the difference between Muse Wonder and Muse Wonder Full?

Full is the larger, more capable version — but it’s locked to Muse Code subscriptions ($50–$200/mo) with no public API. Plain Wonder is available via API at $4/$20.

Is Muse Wonder better than Claude Opus 5.5?

On independent benchmarks, Opus 5.5 still leads (58 vs 56 on the AA index). Wonder wins some agent/coding tests. It’s close, not decisive.

How much does Muse Wonder cost per task?

Artificial Analysis estimates ~$4.52 per Intelligence Index task — about 3x the previous Muse Spark 1.3.

Prices change. Always confirm on Meta’s official documentation before budgeting — this article was verified on October 16, 2026.

Leave a Reply

Your email address will not be published. Required fields are marked *