Meta Muse Wonder explained: Meta's first frontier model, its price and how it compares

On 12 October 2026, Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full, the first Meta models built to compete with the best from OpenAI, Anthropic and Google. They follow the lightweight Muse Spark and the open-weight Muse Spark Open, and they are the first frontier releases since Meta reorganised its AI work under Alexandr Wang in August 2025. Unlike Meta's Llama-era models, these are closed: you use them through Meta's API or its coding app, not by downloading weights.

Wonder and Wonder Full: what's the difference?

How it compares

Meta's own model card says both models score comparably to Claude Fable 5.1 and GPT-6 Astra, and that Muse Wonder beats them on MemoryBench, a test of remembering and using information over time. Independent leaderboards put it firmly in the top group, though not on top.

Price: the expensive end of the top tier

At $20 per million output tokens, Muse Wonder costs about twice as much as the models it ties with. TechCrunch put it roughly level with OpenAI's and Anthropic's most expensive tiers. Cached input at $0.20 helps for agents that resend the same long context. Unless its memory features clearly help your use case, the cheaper models in our October comparison give more for the money.

Active memory and long tasks

The models were trained on Prometheus, Meta's gigawatt-scale cluster in New Albany, Ohio, which Meta described publicly this week. It connects hundreds of thousands of GPUs, with a target of around a million for future Muse training.

How to try it

Create a key at Meta's developer site, then point any OpenAI-compatible code at the Meta Model API.

from openai import OpenAI

client = OpenAI(base_url="<Meta Model API base URL from dev.meta.ai/docs>",
                api_key="YOUR_META_KEY")
reply = client.chat.completions.create(
    model="muse-wonder-1.0",
    messages=[{"role": "user",
               "content": "Explain this SQL query and suggest an index: ..."}],
)
print(reply.choices[0].message.content)

Compare it with the rest of the field in the best AI models in October 2026, and see Meta's open model in Muse Spark Open explained.

Sources: Meta's "Put your ideas to work with Muse Wonder and Muse Wonder Full" developer post; TechCrunch, "Meta releases its first frontier model, Muse Wonder" (12 October 2026); Artificial Analysis Intelligence Index (19 October 2026); Arena text leaderboard (15 October 2026). Vendor benchmarks are Meta's own.

Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 21 October 2026. About me