Meta
Meta prices Muse Wonder at $4 and keeps Muse Wonder Full off the API
Meta released Muse Wonder and Muse Wonder Full, its first frontier models since an AI overhaul, with the API priced at $4 and $20 per million tokens.
Meta released Muse Wonder and Muse Wonder Full, two models it calls “our most capable models to date.” Meta priced the API version at the level it compares to top models from OpenAI and Anthropic. The larger Full model is not offered through the API. Meta says both build on the groundwork of Muse Spark.
Meta announced the pair Oct. 12, 2026, in its Introducing Muse Wonder and Muse Wonder Full post. It is the company’s first frontier release since an AI overhaul that Meta says began 14 months ago.
Muse Wonder Costs $4 Input and $20 Output per Million Tokens
Muse Wonder is priced at $4 per million input tokens and $20 per million output tokens on the Meta Model API, TechCrunch reported Oct. 12. Meta says the cost is roughly equal to GPT-6 Sol or Claude Opus 5.5, according to the same report. OpenRouter’s model list showed the same prices that day, with cached input at $0.20 per million tokens.
The model reads text, image, video and audio and writes text. Meta’s model card gives a context window of 1M tokens.
Muse Wonder Is Available Today and Full Is Limited to Muse Code Subscribers
Meta made Muse Wonder available Oct. 12 through the Meta Model API and Muse Code. Muse Wonder Full is available the same day to Muse Code subscribers only. TechCrunch reported it is not served through the API. Meta said Muse Wonder will soon power experiences in the Muse app, starting with subscribers.
Meta Built the Models Around Memory and Persistence
Meta designed both models around two features it calls Active Memory and Active Focus. Active Memory carries important context across conversations and retrieves relevant details. Active Focus keeps the model working toward a hard goal, trying other approaches, recovering from errors, asking for clarification and checking in before important actions. Meta said the models target law, finance and health work, plus coding.
Meta said it trained them at Prometheus, its gigawatt-scale data center, which TechCrunch placed in central Ohio.
The Benchmark Results Are Meta’s Own
Meta said both models reach state-of-the-art results on key benchmarks for personal and professional agents, memory and coding. Its model card compares them at maximum reasoning effort against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5. The evals include Harvey LAB-AA, Vals Finance Agent V2, HealthBench Pro, AutomationBench-AA, FocusBench, SWE-Marathon and an internal composite Meta calls the Personal Superintelligence Index.
TechCrunch reported, citing the model card, that the scores are comparable to Fable 5.1 and GPT-6 Astra. On MemoryBench, Meta said Muse Wonder outscores Fable, Opus and Astra. The model card says Meta’s runs of competing models are best-effort and not necessarily tuned for proprietary models.
Meta Says Safeguards Are Its Strongest Yet
Meta said Muse Wonder has its strongest safeguards to date against chemical, biological, radiological and nuclear (CBRN) and cyber misuse, while keeping frontier cyber capabilities available for defensive use. The work falls under its Meta Superintelligence Scaling Framework. A full report with more safety evaluations “will be available soon,” Meta said.
What We Think
Meta priced Muse Wonder where it says GPT-6 Sol and Opus 5.5 sit. The price asks buyers to choose on capability. OpenRouter lists Muse Spark 1.3 at $1.25 input and $4.25 output per million tokens.
The strongest model is also the one outside developers cannot call. That limits how much of Meta’s work they can test. Every performance claim above comes from Meta. We would wait for outside testing before moving any workload.