How do you use Meta's Muse Spark?

Muse Spark is the first language model from Meta Superintelligence Labs, announced on April 8. You don't download it or pick it from a list. It is the model now behind Meta AI, Meta's assistant, in the Meta AI app and on the meta.ai website. The upgraded features are starting in the United States, and Meta says they will reach WhatsApp, Instagram, Facebook, Messenger and its AI glasses in the coming weeks. Developers can't use it freely yet: Meta offers it only as a private preview to selected partners.

Meta describes the model as small and fast by design, and the web interface will feel familiar if you have used ChatGPT: a chat box, file uploads, and image and video creation in the same place.

Two modes, and when to use each

Meta AI now has an Instant mode and a Thinking mode. Instant answers right away and suits quick questions, rewording and casual chat. Thinking spends longer working through a problem before replying, which helps with planning, math, science and multi-step questions. For complex requests, Meta says the assistant can also split the work among several subagents, smaller helper processes that each take part of the job at the same time and then combine their results.

A simple habit: start in Instant. If the answer is shallow or misses a constraint, switch to Thinking and ask again rather than arguing with the first reply.

Five tasks to try

The first and last rows lean on Muse Spark's main selling point, multimodal perception: it can read images and charts, not only typed text. Meta says it worked with physicians on health answers. Treat them as background for a conversation with your doctor, not a diagnosis, and think before uploading personal documents.

The fourth row uses Meta's own platforms. Meta AI can bring in public posts from people who know an area and show what people are discussing. In TechRadar's hands-on test, it searched Facebook posts on a topic and then played a role using them. That is useful for local color, but posts are opinions, so check facts such as opening hours at the source.

Here is a worked prompt for the planning row, written to get a usable first answer:

I'm planning a 4-day trip in early July for two adults and two children (ages 6 and 9).
Budget: 2,000 dollars total, not including flights.
Limits: no more than 3 hours of driving per day; one rest afternoon.
Please give a day-by-day plan with rough costs, one rainy-day backup, and a list of things I should book in advance.
Mark any price or opening time I should verify.

The last line matters. Like every chatbot, Meta AI can state details confidently that turn out to be wrong, and asking it to flag what needs checking makes those details easier to find.

How does it compare with ChatGPT and Gemini?

On raw ability, early evidence points to a strong but not leading model. The Guardian reports that independent evaluations put Muse Spark close to the top models from Google, OpenAI and Anthropic in language and visual understanding, but behind them in coding and abstract reasoning. On Artificial Analysis's broad index, which combines many benchmark tests, it shared fourth place. Meta's new lab leader, Alexandr Wang, acknowledged rough edges in the model's behavior and said larger versions are on the way.

On everyday use, the differences are more about where each assistant lives. Meta AI's advantage is reach and context from Meta's apps: it will sit inside messaging apps many people already open every day, and it can draw on public posts, creators and shopping inspiration from Meta's platforms. TechRadar's reviewer found it handled playful, constrained creative requests well and produced polished images with correct spelling, but it could not run tasks in the background. ChatGPT and Gemini, by contrast, are mostly assistants you open on purpose, in their own apps.

The fairest way to choose is a side-by-side test. Pick three tasks you really do, such as a photo question, an email draft and a planning problem, and run the same prompt in all three assistants. Judge accuracy first, then usefulness, then speed. For coding or careful reasoning, the published evaluations suggest checking Muse Spark's work closely; for questions about what is in front of you, or what people nearby are saying, it may be the most convenient of the three.