Artificial Intelligence (AI): Meta launches Muse Wonder to compete with OpenAI and Anthropic

Meta wants to get back into the lead race. The group presented on Monday October 12 Muse Wonder and Muse Wonder Full, two artificial intelligence models designed to accomplish complex tasks, from developing an application to processing an insurance file. With this new range, Mark Zuckerberg's company intends to compete more directly with OpenAI and Anthropic.

The launch comes fourteen months after the start of an overhaul of its AI activities. Beyond performance, Meta relies on assistants capable of maintaining the context from one conversation to another and continuing work over time.

Key Points

Meta launches Muse Wonder and Muse Wonder Full, presented as its most efficient models.

Memory between conversations and the execution of long tasks constitute the main arguments of this range.

Muse Wonder is available to developers, while Wonder Full is reserved for Muse Code subscribers.

With Muse Wonder, Meta focuses on memory and complex tasks

Understanding a lease, organizing a claim with an insurer or building an application: Meta wants to expand the role of its assistants beyond one-off response. The new models must be able to use several documents, follow instructions and sequence the steps necessary for a mission.

This ambition is based in particular on Active Memory, a function intended to retain important information between conversations and to find it when it becomes useful. The objective is to avoid the user having to systematically recall their preferences, constraints or the details of a project.

Meta associates it with Active Focus, which designates the model's ability to continue a task despite difficulties. The assistant should try other approaches, correct mistakes and ask for clarification when necessary. The company also claims to have trained it to consult the user before important actions.

IT development occupies a central place in this strategy. Muse Wonder and Wonder Full must follow a project's requirements, modify code, and repeat testing phases. Meta claims top-notch performance, but the results published at the launch remain evaluations presented by the company: they are not enough to establish general superiority over its competitors.

Paid access and an issue of trust for the personal assistant

Muse Wonder is available via the Meta Model API and Muse Code, the group's programming tool. Its announced price is $4 per million input tokens and $20 output, the tokens being the units of text processed or generated by the model.

The Wonder Full version is, for its part, reserved for Muse Code subscribers. Muse Wonder should also soon join the Muse personal assistant, with an initial deployment to subscribers. Meta does not announce a specific date for free access.

This orientation opens an additional monetization route for the group's investments in AI. It also places confidentiality at the heart of the product: an assistant who works from messages, calendars and personal documents accesses sensitive information.