Meta launches Muse Wonder and Muse Wonder Full, aims to challenge OpenAI and Anthropic

Meta unveiled two new artificial intelligence models on October 13, 2026 — Muse Wonder and Muse Wonder Full — presenting them as contenders to advanced systems from OpenAI and Anthropic. The company said the release is part of a broader push to strengthen its position in the global race to build more capable AIs.

Capabilities and deployment

The firm described the models as built for complex, real‑world tasks: processing legal and financial paperwork, interpreting contracts, answering insurance-related queries and assisting in software development. Meta said Muse Wonder will be available to developers through the Meta Model API and the Muse Code toolkit, while the more advanced Muse Wonder Full will be offered to Muse Code subscribers.

Among the features highlighted are an "active memory" that preserves important information from prior conversations for reuse in later tasks, and an "active focus" designed to let the model work on extended, multi-step problems, adapt its approach when hitting obstacles and resume after errors. Meta also said it plans to integrate the new model into its personal AI assistant, initially for paying subscribers.

Company statements noted that internal benchmarks show performance in some tests comparable to leading systems, though those assessments are based on Meta's own analyses. The firm underlined that independent evaluations are still needed to confirm the models' competitive standing.

Meta said the models were developed with attention to cybersecurity and data protection, given that AI assistants may access messages, calendars and documents. The announcement came after heavy investments — described as billions of dollars by CEO Mark Zuckerberg — in computing infrastructure and technical staff intended to narrow the gap with rivals.

Shares of Meta rose by about four percent during the most recent trading session following the announcement.

Photo: press material from the event