Meta Unveils Muse Wonder AI to Drive Personal Superintelligence
Meta launches Muse Wonder and Muse Wonder Full AI models to tackle complex tasks like legal, financial, and coding work.
Meta has officially stepped into a new era of artificial intelligence with the introduction of its most advanced models yet: Muse Wonder and Muse Wonder Full. Designed to move beyond simple chat interactions, these models aim to achieve "personal superintelligence" by tackling complex, multi-step professional and personal assignments.
A Leap Toward Complex Task Automation
Unlike standard AI models that focus on responding to individual prompts, Meta's new models are engineered for sustained work. This capability allows users to navigate intricate processes such as analyzing legal leases, managing insurance claims, or developing sophisticated software applications.
Meta CEO Mark Zuckerberg highlighted the significance of this launch, stating on X that "Muse Wonder is a major leap towards delivering personal superintelligence for everyone."
Superior Performance in Specialized Fields
The new models have demonstrated significant advantages over existing competitors in several high-stakes benchmarks:
- Legal Expertise: In the Harvey LAB-AA legal agent benchmark, Muse Wonder Full achieved a score of 13.1, outperforming both the standard Muse Wonder (10.0) and GPT-6 Astra (8.6).
- Software Development: On the DeepSWE 1.1 benchmark for long-horizon coding, Muse Wonder Full scored 79.0, slightly edging out the standard Muse Wonder at 78.8.
- Business Workflows: In AutomationBench-AA, which tests end-to-end business automation, Muse Wonder Full scored 45.4, compared to 39.1 for the standard Muse Wonder.
Advanced Features: Active Memory and Focus
To handle these long-duration tasks, Meta has integrated two groundbreaking features into the Muse series:
- Active Memory: This allows the AI to retain and recall relevant information across multiple conversations, ensuring continuity in complex projects.
- Active Focus: This feature enables the models to pursue long-term goals, attempt alternative strategies when stuck, and recover from errors autonomously.
Furthermore, the models are designed to act with caution, seeking user clarification before executing critical actions.
Availability and Safety Protocols
Meta has emphasized a strong commitment to safety and security. The models are trained to refuse harmful requests, protect private information, and assist in strengthening cybersecurity. A detailed report regarding safety evaluations is expected to be released soon.
Currently, Muse Wonder is accessible to developers and builders through the Meta Model API and Muse Code. The more powerful Muse Wonder Full is available specifically to Muse Code subscribers. The company plans to expand these capabilities into its mainstream consumer-facing services in the near future.