Meta has released two new language models, Muse Wonder and Muse Wonder Full, to subscribers of its Muse Code service. Developers can also access Wonder through the Meta Model API, creating a path to use the model outside Meta’s coding interface.
The company says the models compete with leading systems from OpenAI and Anthropic, citing internal benchmark results. Those comparisons have not yet been independently established, and reporting notes that their position relative to Google’s newest Gemini system remains unclear.
Meta intends to use Wonder in its consumer-facing Muse agent, beginning with subscribers, and says it will develop specialized versions for broader access. It has not provided a schedule or detailed rollout plan. The release follows Muse Spark and Muse Image as the company expands the product line created by its reorganized Superintelligence Labs division.
For users, the immediate availability is narrower than Meta’s larger agent ambition: Muse Code subscribers can try both models now, and API customers can use Wonder. Claims about future consumer reach and comparative capability remain plans or vendor test results. Independent evaluations, pricing, context limits and production reliability will determine how the models compare in real workflows.