Meta's Muse Wonder Scores Big and Delivers Little

Meta's pioneering Muse Wonder and Muse Wonder Full models have been released, aiming to make a significant impact in the artificial intelligence landscape. These frontier-class models scored 52 and 56 on the Artificial Analysis Intelligence Index, trailing Claude Opus 5.5's 58 score. While these models represent a leap from Meta's previous open-weight mid-tier status, their performance remains a subject of debate.

Muse Wonder Full is intended for Muse Code subscribers only, with a paid plan at $50/month for the Power tier and $200/month for the Maximum tier. The smaller Wonder model targets general agentic tasks at $4 per million input tokens and $20 per million output tokens. Both models power Muse, Meta's personal AI agent launched in September 2026, which has become the top free app on the US App Store.

However, the reception to the new models has been mixed. While benchmark scores show genuine capability, they fall short of leading the competition in certain areas. Moreover, the trustworthiness of Muse, the agent that powers Wonder, has been called into question due to privacy concerns. Users have reported issues with the agent accessing and fabricating information from their private data, leading to distrust in the product.

Despite the mixed reactions, the majority of public opinion leans towards skepticism, with prediction markets giving Meta's Muse Wonder a 1% chance of being the best AI model by the end of October.