Meta's Muse AI Models Challenge OpenAI, Anthropic
Meta has entered the competition with OpenAI and Anthropic in earnest by unveiling new high-performance AI models, leveraging the success of its AI agent ‘Muse’. The company plans to enhance competitiveness by integrating its proprietary advanced models into the Muse service, which has already secured a user base.
Meta announced on the 12th (local time) the release of its new AI models, ‘Muse Wonder’ and the high-performance version ‘Muse Wonder Full’. Both models will soon be integrated into Muse, initially offered to paid subscribers. Meta stated, “These are the most powerful AI models in our company’s history,” adding, “After overhauling our AI strategy 14 months ago and rebuilding our foundation, we have now reached a stage where we are pushing the boundaries of state-of-the-art models.”
The new models feature ‘Active Memory’, a function that remembers key information from past conversations and retrieves it when needed. They focus on breaking down complex tasks into multiple steps, attempting alternative methods if issues arise, and sustaining long-duration workflows.
In performance evaluations released by Meta, the models outperformed competitors. In a long-term software development assessment, Muse Wonder Full scored 57.5, surpassing Anthropic’s Claude Opus 5.5 (46.3) and OpenAI’s GPT-6 Astra (42.5). In a long-term coding ability evaluation, it achieved 79.0, ahead of Opus 5.5 (74.2) and GPT-6 Astra (74.1). For the ability to remember past conversation contexts, Muse Wonder scored 85.6 and Wonder Full 85.5.
Meta had previously been criticized for lagging behind OpenAI and Anthropic in the race for cutting-edge AI models. However, the release of Muse last month shifted the momentum. According to Bloomberg News, Muse ranked first in free apps on the U.S. Apple App Store and Google Play immediately after its launch. Within a month, it recorded 6.6 million cumulative downloads and 1.8 million daily active users. Meta aims to regain the upper hand in the AI competition by combining Muse’s popularity with improved model performance.
