Meta Superintelligence Labs released the first AI model Muse Spark for WhatsApp and Instagram

Meta Superintelligence Labs has launched Muse Spark — the first model in a new series created after a major reorganization of the company's AI division. This is the first significant product result from the laboratory since Zuckerberg invested billions in a complete overhaul of Meta's AI strategy, recruited researchers from OpenAI, Google DeepMind and other leading AI labs, and established a structure aimed at directly competing with market leaders. Currently, Muse Spark is already operating in Meta AI app and on meta.

ai in the US. Over the coming weeks, the company plans to deploy the model to WhatsApp, Instagram, Facebook, Messenger and Meta Ray-Ban smart glasses. In parallel, the company is rolling out to other countries — specific markets have not yet been named, but given Meta's global presence scale, it likely involves Europe, Latin America and Southeast Asia.

The positioning of Muse Spark deliberately echoes Google's approach to Gemini: the model is described as created specifically for Meta's products. In theory, this means the architecture, contextual memory and system integrations are optimized for specific scenarios within the ecosystem — from messenger chats to voice commands through smart glasses. In parallel, the model has been made available to some Meta partners in closed mode, though collaboration terms have not been disclosed.

Muse Spark is the first in a new series of models from Meta Superintelligence Labs, which in itself signals long-term intentions. The company intends to build its own model stack rather than rely solely on Llama and external providers. Previously, Meta positioned Llama primarily as an open foundation for third-party developers — the open-weights strategy was designed to create an ecosystem and competitive pressure.

But for the product layer embedded in services with billions of users, fundamentally different solutions are needed: with strict security requirements, low latency, managed context and compliance with platform policies. The launch occurred against intense competition in the integrated AI-assistant segment. OpenAI is embedding GPT-4o into voice and search, Google is promoting Gemini 2.

5 Pro as a universal model from Android to Workspace, Apple is rolling out Apple Intelligence across its product line. Meta, with its two billion daily active users, has a unique competitive advantage — built-in distribution through WhatsApp and Instagram without needing to convince users to install a separate app. The key question is not whether Muse Spark will appear in the Instagram feed or in the smart glasses voice assistant, but rather how noticeable the changes will be in everyday use.

Photo caption generation, direct message responses, summarization of long chats, personal assistant in AR glasses — each scenario places fundamentally different demands on latency, memory and response style. The fact that Meta decided to build a specialized model instead of an adapted Llama fork speaks to the seriousness of its intentions. For most users, the change is not yet visible: Meta AI in apps still looks the same.

But under the hood, a fundamental shift has occurred — from a strategy of taking open weights and adapting them to a strategy of building a model from scratch for a specific product ecosystem. This is the distinction that separates companies embedding others' AI in their products from those building the full stack independently. Meta has bet on the second path.

Meta did not publish Muse Spark results on external benchmarks and did not disclose technical architecture details. The actual level of the model will be shown by practice: how noticeably AI features in Instagram and WhatsApp improve, which partners decide to build products based on it. The coming months of rollout will be the main indicator of whether the multi-billion-dollar bet on a proprietary AI stack was justified.