Key facts
- Meta released its Muse Wonder and Muse Wonder Full AI models, its first frontier models.
- The models feature 'active memory' to retain and synthesize user information.
- Muse Wonder matches benchmark scores of OpenAI's Astra and Anthropic's Claude Fable and Opus.
- Muse Wonder Full scored 56 on the Artificial Analysis Index, trailing only Claude Opus.
- Muse Wonder scored 59.1 on the Personal Superintelligence Index, outperforming competitors.
- Muse Wonder is available via API for $4/1M input tokens and $20/1M output tokens.
Meta launched its Muse Wonder and Muse Wonder Full AI models on Monday, positioning them as its most capable frontier models to date and intensifying its competition with OpenAI and Anthropic. The social media giant's shares rose about 4% in afternoon trading following the announcement.
Meta stated that the new models would soon power its Muse personal AI agent, initially for subscribers. The more powerful Muse Wonder Full model will be available to subscribers of its AI coding tool, Muse Code. The company highlighted that these releases are part of a significant overhaul of its AI efforts over the past year, aiming to push the frontier of AI capabilities.
Meta's internal testing indicates that Muse Wonder matches benchmark scores of competitors like OpenAI's Astra and Anthropic's Claude Fable and Opus. Muse Wonder Full achieved a score of 56 on the Artificial Analysis Index, placing it second only to Anthropic's Claude Opus. Muse Wonder scored 59.1 on the Personal Superintelligence Index, outperforming models from both OpenAI and Anthropic. A key feature emphasized is the models' 'active memory,' designed to retain and synthesize user-provided information over extended periods.
The Muse Wonder models are accessible via the Meta Model API, with Muse Wonder priced at $4.00 per million input tokens and $20.00 per million output tokens. The successful rollout of Meta's personal AI agent last month has opened a path to potentially monetize its AI investments through subscriptions.