Archive — US — techcrunch
Meta releases its first frontier model, Muse Wonder
Meta has launched its first frontier models, Muse Wonder and Muse Wonder Full, aiming for personal superintelligence with improved complexity handling and active memory features. The models enjoy high benchmark scores and are priced competitively, marking a significant step in Meta's AI strategy since restructuring its AI division.
Key points
- Meta released Muse Wonder and Muse Wonder Full as its first frontier models for advanced AI capabilities.
- The models are larger and outperform previous models like Muse Spark and competing models such as GPT-6 Astra.
- Pricing is set at $4 per million input tokens and $20 per million output tokens for Muse Wonder.
- Wonder Full is exclusive to Muse Code subscribers and not accessible via general API.
- A focus on ‘active memory’ allows these models to retain user information efficiently over time.
- Muse Wonder outperforms on MemoryBench benchmark compared to its peers, reflecting improved continual learning.
- The release follows a restructuring in Meta's AI efforts led by Alexandr Wang.
Why it matters
- The launch reflects the ongoing trend of companies enhancing AI with memory and personalization capabilities.
- In 2025, AI personal assistants are projected to evolve significantly, with active memory being a key feature for user interaction.
- Meta’s competition with OpenAI and Anthropic in the AI landscape showcases a race for advanced and user-friendly AI models.
- Continual learning is becoming crucial in AI development, impacting user engagement and AI utility.
The introduction of Muse Wonder and Muse Wonder Full signifies Meta's commitment to advancing AI with a focus on complexity and memory, positioning it as a strong competitor in the evolving landscape of personal AI technology.