Meta boss Mark Zuckerberg says the firm has made a major leap in the artificial intelligence race with its new Muse Wonder model, which the company claims can match leading systems from OpenAI and Anthropic.
Meta says it has made a major breakthrough in the artificial intelligence race, with Mark Zuckerberg claiming the company’s latest model can now match technology from leading rivals OpenAI and Anthropic.
The chief executive unveiled Muse Wonder, a new AI system developed by Meta’s Superintelligence Labs division, which the company says represents a significant step towards Zuckerberg’s ambition of delivering “personal superintelligence for everyone”.
Meta has spent years positioning itself as a major player in AI, but its models have generally lagged behind the most advanced systems from rivals.
The company has also struggled to produce a model widely regarded as being at the frontier of AI capability.
That may now be changing.
Meta said Muse Wonder is designed primarily as a personal assistant capable of handling complicated tasks that can stretch over several days.
Examples include organising a cross-country move or dealing with an insurance claim that has been rejected.
Such jobs require an AI system to maintain context and make decisions over multiple stages, increasing the risk of errors in less capable models.
Muse Wonder includes “active memory”, allowing it to retain important information and retrieve relevant details as a task progresses.
Meta also said the model is better at maintaining focus when it encounters problems, while being designed to avoid blindly pursuing a goal.
The system will “ask for clarification when needed” and check with users before taking important actions, Meta said.
The safeguards are particularly notable after a series of incidents involving AI agents from major companies behaving unexpectedly or accessing systems without authorisation.
The company said it had also concentrated on resisting “adversarial instructions” designed to circumvent its safeguards and on protecting sensitive information.
The model is expensive to operate, with costs broadly comparable to the most advanced systems offered by OpenAI and Anthropic. Meta said it will eventually use the technology to power more of its Muse agent, which is currently available for free.
The development reflects a wider shift in Meta’s AI strategy.
Rather than simply competing to produce increasingly powerful general-purpose chatbots, Zuckerberg has emphasised AI systems that can become deeply integrated into users’ lives.
The company hopes Muse Wonder will eventually act as a persistent digital assistant capable of remembering personal information and completing tasks on a user’s behalf.
Whether Meta has genuinely caught up with the industry’s leaders remains to be seen, but the new model suggests its substantial investment in AI is beginning to produce results.
Mark Zuckerberg claims Meta has made huge progress in AI race with new Muse Wonder system
