AI Model Report

Reviews · OCTOBER 14, 2026

Meta's Muse Wonder Lands at $4/$20 — and the Stronger Sibling Is Behind a $50 Subscription

Meta Superintelligence Labs priced Muse Wonder at Claude Opus 5.5 parity through the Meta Model API, but gated the better-scoring Wonder Full behind Muse Code subscriptions starting at $50/month.

By Karl Strauchman · Senior model reviewer · October 14, 2026

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on October 12, pricing the API-available model at $4 per million input tokens and $20 per million output tokens, exact parity with Claude Opus 5.5. The headline-grabbing sibling, Wonder Full, isn't on the Meta Model API at all. It sits behind Muse Code subscriptions at $50 and $200 per month.

That split is the story. The pricing announcement is doing the work of positioning Meta as a credible fourth frontier lab alongside OpenAI, Anthropic, and Google, but the model that justifies the "frontier" claim on benchmarks is the one you can't route API calls to.

The Artificial Analysis numbers show why Meta drew the line where it did. Wonder Full scores 56 on the Intelligence Index, tied with Claude Sonnet 5.5 and three points ahead of GPT-6 Astra and Gemini 4 Argon, both at 53. Claude Opus 5.5 still leads at 58. The API-available Muse Wonder scores 52, one point below Astra. On SWE-Marathon, Wonder Full posts 57.5 against Opus 5.5's 46.3; on the Harvey LAB-AA legal benchmark, 13.1 against 4.2. Opus 5.5 holds Terminal-Bench 4.0 at 59.6, with Wonder Full at 57.1.

Artificial Analysis estimates an average Intelligence Index task runs $4.52 on Muse Wonder, $5.98 on Opus 5.5, and $3.26 on GPT-6 Astra. Meta matched Anthropic's sticker price and undercut it on realized cost per task. It didn't match the GPT-6.1 Sol move from October 1, which landed at a fifth of Astra's per-task cost. For the smaller end of the lineup, Muse Spark 1.3 lists at $1.25 per million input tokens and $4.25 per million output tokens.

Meta AI chief Alexandr Wang and research scientist Kavya Kopparapu are the faces of a fourteen-month push that's now, functionally, a subscription-pricing experiment. The CNBC framing is explicit about that: Muse Wonder is a test of users' willingness to pay. The frontier claim lives on the benchmark chart. The business model lives on the $50 tier.

For anyone evaluating AI stack economics for outreach, agent, or content workloads, the practical read is narrower than the launch suggests. The API model you can actually call is a competent peer to Astra and Argon at Opus pricing, with a small realized-cost edge. The version that beats Opus on coding and legal benchmarks is a consumer-subscription product. Those are two different purchases, and the Bloomberg and TechCrunch coverage, by compressing them into one "frontier" headline, obscures which one landed on your invoice.

That's the comparison that matters. Anthropic improved the model behind the same API endpoint. Meta improved the model behind a different checkout page.

Sources