Meta introduces Muse Wonder with frontier AI capabilities

Meta Platforms introduced its new artificial intelligence models, Muse Wonder and Muse Wonder Full, on October 12. The company claims these models perform at a level comparable to other leading AI systems from OpenAI and Anthropic. Meta CEO Mark Zuckerberg stated that this launch marked a significant step for the company in reaching the frontier of AI development.

The models are part of Meta Superintelligence Labs, following a major restructuring of the company's AI operations. According to independent evaluations, Muse Wonder Full has a score of 56 on the Intelligence Index, slightly below GPT-6 Astra's 58 and Anthropic's Claude Opus 5.5's 58. However, it performs better on specific benchmarks, such as DeepSWE software engineering with a score of 79.0.

The new models, tailored for complex professional tasks, software development, and autonomous digital assistants, include features like Active Memory and Active Focus. Active Memory allows the models to retain relevant information across conversations, while Active Focus enables them to pursue objectives through multiple steps and recover from errors.

These capabilities aim to support tasks from interpreting legal documents to developing software applications. Muse Wonder is accessible via the Meta Model API and Muse Code, while Muse Wonder Full is available only to subscribers of Muse Code. Pricing for the standard model is $4 per million input tokens and $20 per million output tokens.

Muse Wonder will also be integrated into Meta's Muse personal assistant for subscribers, focusing on handling more demanding tasks that require continuity between conversations and external services.