Meta has introduced two new AI models, Muse Wonder and Muse Wonder Full, signalling a more ambitious move into frontier AI. The releases build on Muse Spark but target more demanding tasks, including coding, while placing greater emphasis on long-term memory and personalised assistance.
Larger Models Target More Demanding Tasks
Unlike Muse Spark, which was designed for low latency and lightweight use, Muse Wonder and Wonder Full are significantly larger and, according to Meta, more capable. The company reports stronger benchmark performance and improvements on demanding workloads, including coding.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
According to Meta’s model card, both models achieved benchmark results comparable to Fable 5.1 and GPT-6 Astra in the company’s internal testing. Meta described the release as “the next step on our scaling ladder and a major leap toward personal superintelligence” in its announcement.
Pricing Reflects Meta’s Frontier Ambitions
The models’ pricing places them in the premium AI segment. Through the Meta Model API, Muse Wonder will cost $4 per million input tokens and $20 per million output tokens, putting it in a similar pricing tier to GPT-6 Sol and Opus 5.5.
Muse Wonder Full, by contrast, will not be offered through the API. Meta says it will be exclusive to Muse Code subscribers, tying access to its most advanced model to a premium product. Both models will also power the Muse agent.
Long-Term Memory Takes Centre Stage
Meta is increasingly positioning its AI strategy around assistants that can support users over time rather than respond only to individual prompts. A central part of that approach is what the company calls “active memory”: the ability to retain, organise and synthesise information provided by users across extended periods.
Meta points to the MemoryBench benchmark, which evaluates systems designed for continual learning and memory construction. The company says Muse Wonder now outperforms Fable, Opus and Astra on this measure.
“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,” research scientist Kavya Kopparapu said in a launch video.
New Models Follow Meta’s AI Reorganisation
The launch is also significant within Meta’s broader AI strategy. It is the company’s first frontier-model release since Scale AI founder Alexandr Wang reorganised its AI operations in August 2025. Meta has since introduced Muse Spark and the Muse agent, while making fewer high-profile model announcements than competitors such as OpenAI and Anthropic.
The new models were trained at Meta’s Prometheus data centre in central Ohio, which came online earlier in 2026, according to the company. The facility highlights the growing importance of infrastructure in the race to develop advanced AI systems, where performance depends not only on algorithms and data but also on computing capacity, data-centre investment and access to capital.
As competition intensifies, Muse Wonder and Wonder Full represent a broader bet on larger models, persistent memory and premium AI products. Their longer-term significance will depend on how their reported capabilities translate into real-world performance and whether users find enough value to pay for access.
