Meta Debuts Muse Spark, its First Model From Superintelligence Labs
Meta is also updating its AI assistant with new features, including the ability to run multiple “sub-agents” simultaneously.
Topics
News
- Meta Debuts Muse Spark, its First Model From Superintelligence Labs
- Gen Z, Quick Commerce, AI to Drive India E-Retail Boom
- Air India Starts Search for CEO After Campbell Wilson Resigns
- Anthropic Poaches Microsoft AI Executive for Infrastructure Role
- Uber Taps AWS Chips for Ride Matching, AI Training
- OpenAI Urges Grid Upgrades, Stronger Safety Nets for AI Age
Meta Platforms has introduced a new AI model, Muse Spark, marking the start of a broader series of models being developed by its Superintelligence Labs.
The company says the model is designed to be smaller and faster, while still capable of handling more complex reasoning tasks across areas like science, math, and health. Muse Spark will now power the Meta AI assistant across its app and web platform.
Alexandr Wang, chief AI officer at Meta, shared on X, “Today we’re releasing Muse Spark, the first model from MSL. Nine months ago, we rebuilt our AI stack from scratch. New infrastructure, new architecture, new data pipelines. Muse Spark is the result of that work, and now it powers Meta AI.”
Alongside the model, Meta is updating its AI assistant with new features, including the ability to run multiple “sub-agents” at once. The idea is to break down a single query into parallel tasks, for instance, planning different parts of a trip simultaneously, before combining the results into one response.
The assistant is also being upgraded with stronger multimodal capabilities, allowing it to interpret images in addition to text. This includes use cases like analyzing food items from photos or comparing products visually. Meta says it has worked with physicians to improve how the system responds to common health-related queries, though such use cases remain sensitive and typically require caution.
Another addition is a shopping-focused mode that surfaces recommendations based on trends, creators, and content across Meta’s platforms. It also pulls in public posts and local context when users search for places or topics.
The rollout of these features has begun in the US, with plans to expand to other countries and integrate them across platforms such as Instagram, Facebook, Messenger, and WhatsApp.
Meta plans to offer limited API access to the model for select partners, with the possibility of open-sourcing future versions.
Early reactions on X have been mixed. One user wrote, “In a podcast I listened to today, Yuandong Tian mentioned that in companies like Meta, many people would create models that are designed to look better when the data is presented, but these models don’t perform well in actual use.”
Another post compared the model unfavorably to rivals, saying, “From the benchmarks, it seems @AnthropicAI Mythos is better than Muse Spark. Right?”—a reference to competing models from Anthropic.
A third reaction summed up a more tempered view: “Finally, Meta AI went from being REALLY bad to just ok.”
