Meta is launching two new artificial intelligence models that it says can compete with the industry’s most capable systems. The Verge reported on October 12 that Muse Wonder and Muse Wonder Full are available immediately to Muse Code subscribers, while Wonder is also accessible through the Meta Model API.

The release is a significant test of Meta’s rebuilt AI program. The company had already introduced Muse Spark, the first model from its reorganized Superintelligence Labs division, and followed it with the Muse agent. Wonder and Wonder Full now add a more ambitious claim: that Meta has moved back toward the technical frontier rather than relying mainly on the reach of its consumer products.

Layered AI model blocks rise from a broad network toward a high-performance summit.
Meta says its internal tests place the Wonder models alongside leading rival systems.

Meta says its internal evaluations place the new models alongside leading systems from OpenAI and Anthropic, including GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Those comparisons remain company claims rather than an independent verdict in The Verge’s report. The publication also said the models’ standing against Google’s Gemini 4 Argon is unclear because that system’s capabilities are still being assessed.

The company plans to use Wonder to power its Muse AI agent, beginning with subscribers. It also intends to create specialized versions that can eventually reach a broader Muse audience. Meta did not provide a timeline or detailed rollout plan for either step, leaving the scale and pace of that integration uncertain.

An abstract AI core powers a task agent that branches into multiple work paths.
Meta plans to make Wonder the engine behind its Muse agent, beginning with subscribers.

That planned connection between model and agent could matter as much as benchmark scores. Muse is designed to perform tasks for users, so a stronger underlying model could affect how well it plans, reasons and completes multi-step work. For Meta, the combination also creates a route from a developer-facing model launch to a product distributed across its larger ecosystem.

The launches follow billions of dollars in spending on AI talent and infrastructure, according to The Verge. Meta has also been widening the Muse lineup with Muse Image and has emphasized consumer AI agents as a central part of its strategy. Wonder and Wonder Full therefore arrive as part of a coordinated push, not a standalone research release.

The immediate question is whether outside testing will support Meta’s internal results. Availability through Muse Code and the Meta Model API gives developers a path to start forming that judgment, but the report did not include independent performance data. Meta has made its frontier claim; the next evidence will come from how the models perform beyond the company’s own benchmarks.