Meta has officially launched a new series of AI models dubbed 'Muse Wonder,' which the company asserts has caught up to leading competitors like Anthropic and OpenAI. CEO Mark Zuckerberg described the new models as a 'much larger and more advanced class' that brings Meta to the AI frontier. The Wonder series consists of two versions: Muse Wonder, a smaller model for daily tasks, and Wonder Full, a larger, more capable variant. Internally codenamed 'Watermelon,' the project was teased by Meta employees before its release. This launch aims to validate Meta's massive AI investment, which had previously drawn skepticism from investors due to the hundreds of billions of dollars spent on the technology.
Meta has released its latest AI models, Muse Wonder and Muse Wonder Full, claiming they incorporate 'active memory' and 'active focus' features that enable the models to retain context across conversations and maintain focus on tasks. The models are said to excel at legal, financial, and health-related tasks and deliver frontier coding performance. Meta plans to integrate Wonder into its personal agent Muse for paid subscribers, with pricing tiers of $4 per million input tokens, $20 per million output tokens, and subscription options of $50/month and $200/month. CEO Mark Zuckerberg highlighted the significant compute, investment, and scientific effort behind the development, and Meta expects to spend up to $145 billion this year on AI.
Meta releases its first frontier model, Muse Wonder https://t.co/2mTYv6k6So
Meta is aggressively expanding its AI portfolio by launching two new models, Muse Wonder and Muse Wonder Full. These models are currently available to Muse Code subscribers and via the Meta Model API. Meta claims these models rival industry-leading offerings like OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5 and Claude Fable 5.1, based on internal benchmarks. The company intends to integrate these models into its Muse AI agent in the future, though no specific timeline has been provided. This move is part of Meta's broader strategy to regain a competitive position in the AI sector through significant investment in infrastructure and talent.
Meta says its new models have active memory and active focus abilities.
Mark Zuckerberg said 'getting to this point both required a large amount of compute and investment, and a lot of very rigorous scientific work.' Meta says Muse Wonder Full boasts frontier capabilities. Meta says the largest version outperforms Anthropic’s second-best model, Claude Fable 5.1. Meta says the model outperforms OpenAI’s best model, GPT-6 Astra. Meta says it eventually plans to power its Muse AI agent with Wonder. Meta is spending hundreds of billions of dollars on AI. Meta says its new models have active memory and active focus abilities. Meta AI chief Alexandr Wang had told staff that Watermelon had caught up to OpenAI. Meta says the models are adept at navigating legal, financial and health-related tasks.
Meta's first frontier model is out. Muse Wonder, codenamed Watermelon. https://t.co/3h1rpU0U9j
Meta says it plans to bring Wonder to its personal agent Muse, beginning with paid subscribers. The company intends to integrate these models into its Muse AI agent in the future, though no specific timeline has been provided. This move is part of Meta's broader strategy to regain a competitive position in the AI sector through significant investment in infrastructure and talent.
On 2025-05-22, Meta launched Muse Wonder and Muse Wonder Full AI models. On 2026-10-12, Meta released Muse Wonder and Muse Wonder Full AI models and Mark Zuckerberg posted about the launch on X.
Meta releases its first frontier model, Muse Wonder https://t.co/0fs7kWin4u