Meta Muse Spark debuts: New AI model to challenge Google, OpenAI

Meta on Wednesday unveiled Muse Spark, a new natively multimodal large language model developed under Meta Superintelligence Labs (MSL), and made it available through the Meta AI app and the meta.ai website. The company says the model narrows performance gaps with leading competitors and will be rolled out across Facebook, Instagram and WhatsApp over time.

According to Meta's disclosures, Muse Spark—internally codenamed Avocado—was built in roughly nine months by a team led by Alexandr Wang and represents a step up from the Llama 4 family. The model accepts voice, text and image inputs while producing text outputs, offers a fast mode for casual queries and several dedicated reasoning modes, and includes product-focused features such as a shopping mode that ties model outputs to user interests and behavior. Meta also indicated plans to publish a version under an open-source license.

The announcement follows Meta's high-profile investment and talent acquisition strategy—including the company's investment in Scale AI and the hiring of Wang to head its superintelligence effort—which restructured Meta's foundational model and research teams and signaled an accelerated push to close the gap with rivals. Industry coverage has framed Muse Spark as the first tangible product from that reorganization.

From a market and commercial perspective, Muse Spark is not a direct revenue product at launch, but its integration into Meta's family of apps could materially affect ad targeting, recommendation quality and engagement metrics over time. Meta emphasises features that can drive differentiated ad experiences and commerce flows; if these capabilities translate into higher engagement or ad effectiveness, they could support upward revisions to monetization forecasts.

Analysts characterize Muse Spark as competitive on certain multimodal and domain-specific tasks but not a wholesale leap to state-of-the-art across the board. Near-term investor focus will likely center on how quickly Meta operationalizes Muse Spark across products, the pace of adoption, and regulatory or privacy-related constraints tied to model data use. The model's eventual commercial impact will depend on execution and measurable improvements in user metrics and advertiser ROI.