Meta has had an AI assistant in WhatsApp, Instagram, and Facebook for over a year. Few people use it by choice. Muse Spark is the company's bet that the problem was model quality.
The new model is the first release from Meta Superintelligence Labs, the organization Mark Zuckerberg built after overhauling Meta's AI teams and committing billions in additional investment. As of today, Muse Spark powers the Meta AI app and website in the US. WhatsApp, Instagram, Facebook, and Messenger get the upgrade in the coming weeks, according to Meta's announcement.
A Distinct Move from Llama
Muse Spark is a separate product from Meta's Llama series - the open-source models developers have been downloading and deploying since 2023. Llama 3 was widely praised for its performance relative to its size and became the most popular base model for researchers building specialized AI tools. Muse Spark appears to be Meta's consumer-facing counterpart: a model designed to be experienced through Meta's apps, not downloaded and modified.
Meta hasn't published detailed benchmarks alongside the announcement, which makes direct comparisons to ChatGPT, Claude, or Gemini difficult. What we know is that Muse Spark now runs Meta AI - a product that previously ran on a mix of Llama variants and wasn't competitive on tasks like nuanced writing, reasoning, or long research queries.
Distribution Advantage vs. Engagement Problem
The reach is hard to overstate. WhatsApp has roughly 2 billion monthly active users. Facebook has 3 billion. Messenger and Instagram add hundreds of millions more. No other AI company has this kind of default install base. ChatGPT users had to download an app or visit a website. Meta AI users just... already have it.
The problem has been that being present isn't the same as being chosen. Meta AI has been criticized for inserting itself into group conversations uninvited and for responses that feel generic compared to ChatGPT. If users don't find it useful, the distribution advantage produces only impressions, not engagement.
The theory behind Muse Spark is that a significantly better model, shipped through the same surfaces, finally earns the attention Meta AI has been handed by default. Whether it's actually that model won't be clear until independent evaluations are published.