Meta's AI Models Challenge OpenAI and Anthropic

Meta Superintelligence Labs (MSL) introduced Muse Wonder and Muse Wonder Full on October 12, 2026, positioning them as strong contenders against industry leaders like OpenAI and Anthropic. These advanced AI models aim to excel in complex software engineering, legal work, business automation, and long-running agent tasks.

Independent evaluations by Artificial Analysis show Wonder Full scoring 56 on the Intelligence Index, matching Anthropic's Claude Sonnet 5.5 and outperforming models like GPT-6 Astra and Gemini 4 Argon. The standard Muse Wonder achieved a score of 52. Unlike previous Llama models, the Wonder series is not open-weight, with access offered via a commercial API for the standard model and paid subscriptions for Wonder Full through Meta's Muse Code assistant.

The Wonder models are designed for sophisticated agentic tasks, capable of navigating multiple applications and retaining information over extended periods. They accept multimodal inputs including text, images, audio, and video, and support context windows up to one million tokens. API pricing for the standard Wonder model starts at $4 per million input tokens.

Enterprise adoption faces considerations regarding cost, data governance, and the proprietary nature of the models, which are not available for self-hosting. Meta offers different data-use arrangements, including a Contributor mode for its Muse Code subscriptions, which allows Meta to use submitted data for model improvement. Businesses handling sensitive information must carefully review these terms.