Muse SparkThe short answer
Muse Spark is the reasoning model behind Meta AI experiences. It is native multimodal, which means vision was trained alongside language instead of being attached as a separate service. A screenshot, a chart, a diagram and a paragraph can all be handed over as one context, and the model reasons about the relationships between them.
On top of that, Muse Spark plans. In Contemplating mode several agents reason in parallel and their answers are reconciled before anything is written back. The result is slower than a single-shot response and far more reliable on multi-step science, research and business questions.
What Muse Spark can do
- Understand mixed inputs. Upload an image, a chart or a diagram and ask questions about it in the same sentence as the rest of your context.
- Reason in parallel. Contemplating mode fans a hard question out to multiple agents, then merges the answers into one plan.
- Create artefacts. Images with reasoning behind them, custom websites and landing pages, dashboards and mini-games — produced, not just described.
- Personalise. Recommendations surface across Meta apps, and visual understanding powers Meta AI glasses experiences.
How it differs from a chat model
Chat models answer. Muse Spark plans and then produces. The difference shows up in the shape of the response: instead of a paragraph describing a landing page, you get the page; instead of instructions for a figure, you get the image with the assumptions it was built on.
That also changes how you prompt it. Describe the outcome, attach the material you already have, and say what the artefact is for. The model handles the decomposition.
Contemplating mode, explained
Contemplating mode is the deliberate path. Several agents approach the problem from different angles — definitions, constraints, counter-examples, sources — and the model reconciles them before answering. Use it for anything where being fast matters less than being right: research summaries, technical explanations, pricing and policy analysis.
Plans, credits and access
Access to the composer on this site runs through a Google sign-in and an active subscription. Every plan includes multimodal reasoning, image generation and code creation; higher tiers add Contemplating mode, higher generation limits, priority queueing and API access. The current prices, credit allowances and annual discount are listed on the pricing page.
Technical detail — capabilities, reasoning modes and how credits are spent — lives in the documentation.
Common questions
What is Muse Spark?
Muse Spark is Meta's native multimodal reasoning model. It reads text, images and charts in one context, plans with multiple agents, and creates images, websites, mini-games and dashboards from a single prompt.
How is Muse Spark different from other models?
Vision and reasoning were trained together rather than stitched on afterwards, so a chart, a screenshot and a question can be handed over as one problem. Contemplating mode adds parallel agents for harder questions.
What can Muse Spark create?
Images with reasoning behind them, custom websites and landing pages, dashboards, mini-games, plus written explanations, summaries and step-by-step plans.
What is Contemplating mode?
A slower, more deliberate path: several agents reason in parallel and their answers are reconciled before Muse Spark responds. It is the mode for multi-step science, maths and research questions.
Is Muse Spark available in Meta AI?
Muse Spark is the reasoning model behind Meta AI experiences, including visual understanding on Meta AI glasses and personalised recommendations across the Meta apps.
Who is Muse Spark for?
Teams and individuals who work with mixed inputs — analysts reading charts, founders writing product briefs, students and researchers working through multi-step problems, and creators producing images and sites.