[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space Latent Space ● Covered by 29 sources
Meta shipped Muse Wonder, a new frontier model family. Wonder Full beats some rivals, but the pricing and access split look messy.
Based on reporting by Latent Space, Latent Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta’s Superintelligence Labs finally dropped Muse Wonder, code-named Watermelon, and the pitch is simple: this is the step that puts the lab at the frontier. The timing matters. The lab has been at it for roughly 14 months, and Meta is now trying to sell Muse not as a one-off model, but as a whole family built for agent work.
There are two versions. Muse Wonder is the faster one and is available through the Meta Model API and Muse Code. Muse Wonder Full is the max-reasoning tier, locked to Muse Code subscribers. Meta is also leaning hard on product framing here: personal and professional agentic work, with claims around memory over days and months, staying focused through blockers, and stronger performance in finance, legal, and health.
The benchmark picture is strong, but not clean enough to pretend this is settled. Artificial Analysis puts Wonder Full at 56 on its Intelligence Index and Wonder at 52, while Opus 5.5 sits at 58. Wonder Full also improves on all ten index evals, with the biggest jumps showing up in agentic work: Terminal-Bench, GDPval-AA, and AA-Briefcase all move noticeably. Meta’s own headline numbers include 57.5% on SWE-Marathon and 79.0% on DeepSWE v1.1 for Wonder Full.
And then there’s the part Meta will probably wish had gone smoother: the launch-day quota bug. Both models were burning through Muse Code usage faster than expected, and the fix came the same day with a quota reset. Pricing is also not exactly subtle, with leaked numbers pointing to $4 per million input tokens and $20 per million output tokens, plus a 1 million context window. Wonder Full’s token price still isn’t published, which leaves the flagship model looking very capable and still a little awkward to buy into.
My take — AI-written commentary, not fact-checked reporting
Meta keeps trying to turn model launches into a declaration of inevitability, and this one at least has the evals to back up the swagger. But the access split, missing price for the top tier, and launch-day quota mess are classic frontier-lab chaos. Big model smell, indeed: impressive, expensive, and still held together with duct tape at the edges.
Read more about this at: Latent Space