Meta Superintelligence Labs released Muse Wonder and a larger sibling, Muse Wonder Full, on 12 October 2026, calling them the company's first frontier-class models. Both outscore the lightweight Muse Spark on Meta's own benchmarks, particularly for coding and multi-step tasks, and both now power Meta's Muse agent.
“We put a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond”
Kavya Kopparapu, Meta research scientist, in a launch video · techcrunch.com ↗
Access differs between the two. Wonder is sold through the Meta Model API at $4 per million input tokens and $20 per million output tokens, roughly the price of GPT-6 Sol or Claude Opus 5.5. Wonder Full is reserved for Muse Code subscribers and is not offered through the API. Wonder takes text, images, video and audio in a one-million-token context window, and the models were trained at Meta's Prometheus data center in central Ohio.
Meta's launch material for the Muse Wonder models, as pictured by TechCrunch
Meta · Image: TechCrunch
Meta's tests put the pair level with Fable 5.1 and GPT-6 Astra, and Wonder tops MemoryBench against Fable, Opus and Astra. On Meta's chart Wonder Full scores 79.0 on DeepSWE 1.1 against about 74 for Opus 5.5 and Astra, and 57.5 on SWE-Marathon against 46.3 for Opus 5.5. Opus 5.5 still leads on HealthBench Pro, OSWorld, Terminal-Bench and ProgramBench. On the independent Artificial Analysis index it scores 58, ahead of Wonder Full at 56 and Wonder at 52. Meta stresses "active memory" and "active focus" as the features that set the models apart.
About Meta Superintelligence Labs
Meta Superintelligence Labs is the artificial intelligence division of Meta Platforms, based in Menlo Park, California, and founded on 30 June 2025. It works on artificial superintelligence and makes the Muse family of generative AI models.
It followed Meta's purchase in June 2025 of a 49% non-voting stake in Scale AI for more than $14 billion, and Alexandr Wang became chief AI officer. In August 2025 the division was reorganised into four groups, and on 20 November 2025 chief AI scientist Yann LeCun left to found Advanced Machine Intelligence Labs. In August 2026 it released Muse Glimmer, a 30-billion-parameter open-weights model under the Apache licence.
1
Models indexed
Models on AIPopDB
This is the first top-tier release since Alexandr Wang, the Scale AI founder, restructured Meta's AI work; TechCrunch dates that to August last year, while ProPakistani gives no date. Meta had since shipped only Muse Spark and the agent. Meta describes the release as the next step on its scaling ladder toward personal superintelligence.
Sources · 2