Meta says Muse Spark will power Meta AI’s “shopping mode” and could be partly open source

April 8, 2026
A smartphone with a shopping cart depicting the concept of online shopping in a colorful studio setup.
Photo by Nataliya Vaitkevich on Pexels

What Meta announced

Meta on Wednesday unveiled Muse Spark, a new in-house AI model it says closes much of the performance gap with OpenAI, Anthropic and other leading labs. The model — code‑named Avocado and developed over the past nine months by a team led by Alexandr Wang — is described by Meta as a major step up from its Llama 4 family. Meta says Muse Spark will immediately handle queries in the Meta AI app and on Meta.ai.

How it works and where it will run

Muse Spark reportedly uses a fast mode for casual, low-latency questions and multiple reasoning modes for harder tasks. Meta says the model will roll out beyond the app and website into Facebook, Instagram and WhatsApp, and that it powers a new “shopping mode” feature intended to help users find and buy items. It has been reported that Muse Spark is competitive with the latest models on specific tasks such as multimodal understanding and processing health information — though a Meta executive told Axios it’s not claiming a new state of the art.

Claims, caveats and a privacy flag

Consumers should take those claims with a grain of salt. Meta’s privacy policy gives the company broad latitude in how it uses data shared with its AI systems, and that matters when shopping histories, pictures and personal questions are involved. A Black Mirror–adjacent worry? Sure. But this is a familiar industry trade-off: convenience and capability versus control and transparency.

Why this matters

Meta frames Muse Spark as one step toward a vision of personal superintelligence — personal assistants that know you, help you shop, and triage information. It has been reported that Meta also plans to release a version of Muse Spark under an open‑source license, a move that would feed the broader trend toward more accessible, competitive AI tooling. So the race continues: will openness and scale reshape who controls the next wave of AI, or will the biggest platforms hold the reins?

Sources: axios.com