Meta has launched Muse Wonder and Muse Wonder Full, two artificial-intelligence models it describes as its most capable systems and a step toward Mark Zuckerberg’s vision of “personal superintelligence.” The models are aimed at long-running personal and professional tasks, coding, and computer-use agents that can operate across browsers, desktops and mobile applications.
Meta says Muse Wonder’s Active Memory carries important user context between conversations, while Active Focus helps an agent maintain a goal, adjust its approach when it encounters obstacles and continue working on difficult assignments. The models can plan lengthy tasks, delegate work to sub-agents, use tools in parallel and recover from errors. Both have a context window of 1 million tokens and accept text, image, video and audio inputs, with text output.
Independent Analysis recorded Intelligence Index scores of 56 for Muse Wonder Full and 52 for Muse Wonder at maximum reasoning effort. Muse Wonder Full ranked ahead of Claude Fable 5. 1 and GPT-6 Astra, both at 53, but below Claude Opus 5. 5 at 58. The evaluation also found that Muse Wonder Full placed in the top three on six of 10 tests, including first place on CritPt and GDP.
The same testing showed a substantial improvement over Meta’s Muse Spark 1. 3, which scored 48 on the index. Artificial Analysis also found knowledge performance remained weaker than that of several leading rivals, even as the model performed strongly on agentic and coding evaluations.
Access is divided between the two releases. Muse Wonder is available through Meta’s first-party Model API and Muse Code, while Muse Wonder Full is restricted to Muse Code’s Power Usage and Maximum Usage subscriptions, priced at $50 and $200 per month. Muse Wonder’s API pricing is $4 per million input tokens, 20 cents per million cached input tokens and $20 per million output tokens.
The launch puts Meta’s large computing investment and newly assembled AI research operation under direct comparison with OpenAI and Anthropic. Meta Superintelligence Labs head Alexandr Wang called the models frontier systems, while Meta has not said whether the Wonder family will be open-weight. AI, disputed the performance claims on X and said her company would publish detailed benchmark results.
Meta says Muse Wonder will also power future experiences in its Muse assistant, initially for paid subscribers. The company’s emphasis on persistent memory and agents acting on a user’s behalf also brings privacy and security considerations into the product’s rollout, while independent testing will determine how consistently the claimed advantages hold beyond the published evaluations.