Meta has launched Muse Wonder, its latest artificial intelligence model, as the company steps up efforts to compete with leading AI developers OpenAI and Anthropic. The technology is designed to power AI assistants and could cost users between $20 and $100 a month, depending on the subscription tier.
Meta Chief Executive Mark Zuckerberg described the new system as a major advance in the company’s AI development. “Muse Wonder is a much larger and more advanced class of models that brings us to the frontier,” he wrote on X on Monday.
The company said the Muse Wonder family includes two versions: Muse Wonder, a smaller model intended for everyday tasks, and Wonderful, a more capable system designed for demanding applications.
According to Meta, the models can assist with work in areas such as law, finance and healthcare, while also offering advanced coding capabilities for developers building games, applications and businesses.
Meta has also highlighted the models’ memory functions. Its active memory technology is designed to help AI assistants retain relevant details from previous interactions and understand tasks over time, allowing them to provide more consistent responses.
Meta Superintelligence Labs head Alexandr Wang described the release as the company’s most advanced model so far, saying it “rivals the big (boys)” in a post on X.
The company published benchmarking data from Artificial Analysis, an independent AI evaluation firm, that appeared to show the Wonder models outperforming Meta’s earlier Spark system and several models developed by OpenAI, Anthropic and Google. However, Artificial Analysis also ranked Wonder among the more expensive models available.
The launch follows a substantial investment by Meta to strengthen its position in the AI industry. In June 2025, the company invested $14.3 billion in Scale AI and recruited its co-founder and chief executive, Wang, to lead its foundational model development efforts.
Zuckerberg subsequently recruited executives and researchers from competing firms, including OpenAI, Anthropic and Google.
Meta introduced its Spark model in April, moving away from the open-source approach associated with its Llama family. However, Spark reportedly struggled to match some competing systems, increasing pressure on the company to deliver stronger results.
The latest release follows other AI products introduced by Meta this year, including Muse Code, a coding tool launched in August, and an agentic application in September. Agentic AI systems can carry out tasks such as research, shopping and booking reservations with limited user intervention.
At its Connect developer conference last month, Meta also unveiled Muse Charm, a handheld AI device intended to let users interact with its assistant independently of a smartphone.
The company’s expansion comes amid growing debate over AI safety. Zuckerberg joined executives from Google, OpenAI, Anthropic, xAI and Nvidia in signing a pledge proposed by US President Donald Trump to strengthen safety measures before releasing advanced models.
Some industry leaders have called for more cautious development following incidents involving AI systems acting outside testing environments or taking unauthorised actions.
Anthropic chief executive Dario Amodei told the United Nations last month that his company would slow development as much as necessary to ensure its successive AI technologies were safe.
Meta’s challenge will be to demonstrate that Muse Wonder can compete on performance while addressing growing concerns about the risks posed by increasingly capable AI systems.
