--- title: 'Muse Spark' description: "Meta Superintelligence Labs' multimodal reasoning model family for agentic and coding work, separate from the Muse agent and Muse Code applications." canonical_url: 'https://darkfactory.dev/glossary/muse-spark' markdown_url: 'https://darkfactory.dev/glossary/muse-spark.md' collection: glossary date_published: '2026-10-07T00:00:00-04:00' date_modified: '2026-10-07T00:00:00-04:00' --- # Muse Spark ## Definition Muse Spark is Meta Superintelligence Labs' family of multimodal reasoning models for agentic and coding work. The models interpret supplied inputs and generate text, including proposed tool calls. An application supplies the tools, executes allowed calls, and returns results for further reasoning. Spark powers applications such as the Muse personal agent and Muse Code. Each application's harness determines its memory, execution environment, and permissions. ## Origin and attribution Meta introduced Muse Spark on April 8, 2026, as the first model in its Muse program. Development is credited to Meta Superintelligence Labs, without naming a sole creator. The September 2 Spark 1.3 announcement describes improvements to longer agentic tasks and coding, including how it handles changing instructions. Name the evaluated version when comparing results. ## Versions and capabilities The October 7 catalog lists versions 1.1, 1.2, and 1.3 with separate API identifiers. Spark accepts text and several other input modalities while producing text. The documented context window is 1,048,576 tokens; that capacity does not establish reliable recall or instruction following across every long input. The catalog marks audio understanding in 1.3 as not fully supported and warns that response quality may be degraded. It recommends 1.2 or dedicated transcription for audio work. A modality appearing in a table therefore needs its accompanying qualification. Spark 1.3 supports a maximum reasoning effort on its Standard tier. Comparisons need the selected reasoning setting as well as the model identifier: a change in inference compute can change latency, cost, and task performance. ## Data use and deployment limits Meta's API documentation distinguishes Standard access, whose request data is excluded from training, from Contributor variants that exchange training permission for a lower price. Model versions and data-use tiers are separate choices. Check both before supplying private context. A Spark model can propose actions but does not itself possess a user's account access or approval. The Muse product's Sentinel and a coding application's sandbox are harness controls. Their guarantees do not transfer automatically to a different application calling Spark. Meta's release claims and demonstrations describe selected evaluation settings. Test the complete deployment on its intended work, including tool failures and conflicting instructions. Verify external results independently of the model's explanation of what it completed. ## Distinguish it from nearby terms - Muse models is the broader brand entry. Spark is its general reasoning family. - Muse and Muse Code are agent applications that use models. Their persistence and authority come from the deployed harness. - Muse Glimmer is a separately released distilled model with downloadable weights. Spark API terms do not define Glimmer's serving arrangement or license. - Multimodal input does not imply native image generation, speech synthesis, or uniform quality across every accepted format. ## Check your understanding A team switches from Spark 1.2 Standard to Spark 1.3 Contributor for an audio-heavy task. Which modality limitation and data-use change need review before it treats the switch as a routine version upgrade? ## Also called Muse Spark model ## Related terms - [Muse models](https://darkfactory.dev/glossary/muse-model) - [Muse](https://darkfactory.dev/glossary/muse) - [Muse Glimmer](https://darkfactory.dev/glossary/muse-glimmer) - [AI model](https://darkfactory.dev/glossary/ai-model) - [Large language model (LLM)](https://darkfactory.dev/glossary/large-language-model) - [Multimodal model](https://darkfactory.dev/glossary/multimodal-model) - [Reasoning model](https://darkfactory.dev/glossary/reasoning-model) - [Proprietary model](https://darkfactory.dev/glossary/proprietary-model) - [Model distillation](https://darkfactory.dev/glossary/distillation) - [Context window](https://darkfactory.dev/glossary/context-window) - [Model routing](https://darkfactory.dev/glossary/model-routing) - [Agent harness](https://darkfactory.dev/glossary/agent-harness) - [Evaluation (eval)](https://darkfactory.dev/glossary/evaluation) ## Related factory areas - [Model selection, routing & budgets](https://darkfactory.dev/factory/model-routing-budgets) - [Tools & project interfaces](https://darkfactory.dev/factory/tools-interfaces) - [Verification, evaluation & quality truth](https://darkfactory.dev/factory/verification) - [Data, database & irreversible state lifecycle](https://darkfactory.dev/factory/data-lifecycle) ## Evidence and further reading - [Meta: Introducing Muse Spark](https://ai.meta.com/blog/introducing-muse-spark-msl/) - [Meta: Introducing Muse Spark 1.3](https://research.meta.ai/blog/introducing-muse-spark-1-3) - [Meta: Model catalog](https://dev.meta.ai/docs/models)