- Meta released Muse Wonder and Muse Wonder Full on Oct. 12, 2026, calling them its first frontier-class models.
- API pricing for Muse Wonder is $4.00 per million input tokens and $20.00 per million output tokens.
- Wonder Full is limited to Muse Code subscribers and is not served through the API.
- The benchmark claims come from Meta’s own model card, so independent testing still matters.
In a launch video, Meta research scientist Kavya Kopparapu said the team put in a tremendous amount of effort so that your model can remember what is important to you. That sentence is the real pitch behind Muse Wonder. Meta is not only chasing leaderboard scores. It is betting that a model which retains your context over time is worth paying frontier prices for.
What Meta Actually Shipped
Two models, two access tiers
Meta’s Superintelligence Lab released Muse Wonder and Muse Wonder Full on Monday, Oct. 12. Both are larger and more capable than the lightweight Muse Spark, with higher benchmark scores and stronger results on complex tasks such as coding. Both will also power the Muse agent.
Access is split. Muse Wonder is available through the Meta Model API. Wonder Full is reserved for Muse Code subscribers and is not served through the API, so developers who want to build on it directly cannot do so yet.
Trend Insight — Keeping the larger model behind a subscription is a distribution choice, not just a pricing one. It pushes developers toward Meta’s own coding product instead of toward a neutral API.
Pricing and the Benchmark Claims
Priced like the leaders
Muse Wonder costs $4.00 per million input tokens and $20.00 per million output tokens through the Meta Model API. According to TechCrunch, that is roughly on par with GPT-6 Sol and Claude Opus 5.5. Meta is not trying to win on price here.
Scores come from Meta itself
Per Meta’s model card, Wonder and Wonder Full score comparably to Claude Fable 5.1 and GPT-6 Astra. Muse Wonder reportedly beats Fable, Opus, and Astra on MemoryBench, a benchmark for memory and continual learning. These are first-party numbers. Until third-party evaluations appear, treat them as a claim to verify, not a settled ranking.
Trend Insight — The one benchmark where Meta claims a clear win is the one that measures its headline feature. Teams should test memory behavior on their own workloads before paying frontier rates for it.
Why Active Memory Is the Bet
Remembering what you told it
Meta emphasizes what it calls active memory: the ability to retain and synthesize information a user provides over time. In its blog post, the company says the models mark the next step on its scaling ladder and a major leap toward personal superintelligence.
The models were trained at Meta’s Prometheus data center in central Ohio, which came online earlier in 2026. This is the first frontier release since Alexandr Wang restructured Meta’s AI efforts in August 2025. Since then Meta has shipped Muse Spark and the Muse agent, but not a state-of-the-art model until now.
What developers should check
Persistent memory raises practical questions. Where is user data stored, how can it be deleted, and how does memory behave across sessions and accounts? Meta’s launch materials emphasize the capability, so these details are worth confirming in the API documentation before building customer-facing features on it.
Trend Insight — Frontier competition is shifting from raw reasoning scores toward persistence. The model that remembers your context becomes harder to switch away from, which is exactly why vendors are racing to own it.
Related
- Anthropic’s New Model Costs 90% Less. What’s the Catch?
- Satya Nadella Just Said the Quiet Part About AI
- Anthropic Just Pulled the Plug on Its Own AI
- OpenAI’s $70 Billion Number Just Lost $20 Billion
- Mistral’s 1T Model Has One Big Catch
Sources
- TechCrunch: Meta releases its first frontier model, Muse Wonder
- TechCrunch AI category
- Anthropic News (pricing context)
AI Biz Insider · AI Trends EN · aibizinsider.com
답글 남기기