Meta's New AI Takes on GPT-6 and Claude: Muse Wonder Explained

Meet Muse Wonder

Meta has launched Muse Wonder and a bigger sibling, Wonder Full. They are the company's first frontier models, which means Meta is positioning them against the strongest systems from OpenAI and Anthropic.

The focus is agents. Meta says both models are built for AI agents that remember what matters to you and keep going when a task gets hard. That pairs long-term memory with persistence on difficult, multi-step work.

The scores, per Meta

Meta's headline result comes from MemoryBench, an internal evaluation that measures how well a model remembers and uses a user's context across a long interaction history. Muse Wonder scored 85.6%, ahead of Claude Opus 5.5 at 81.5% and GPT-6 Astra at 81.4%.

Notice the gap between the two Meta models is a tenth of a point, so on this test Wonder and Wonder Full are effectively tied. The lead over the other models is roughly four points.

What the independent index says

On the independent Artificial Analysis index, the picture is closer and less favorable to Meta. Claude Opus 5.5 scores 58 and Wonder Full scores 56. Wonder Full is near the top, but it does not take first place.

Put together, the two results suggest Meta has built something competitive that is especially strong at memory, without clearly beating the best models across the board.

How to get it

Access depends on which model you want.

The takeaway

Meta is betting that memory is what separates a useful agent from a forgetful chatbot, and its first frontier models are built around that idea. The benchmark Meta chose puts Wonder in front. The independent index keeps Claude Opus 5.5 narrowly ahead. For developers, the practical news is that you can try Wonder today at a published price.