Muse Spark
The first Muse model, with native multimodal reasoning, tools, visual chain of thought, and multi-agent orchestration.
FAMILY / META
Muse
Meta Superintelligence Labs’ new family spanning multimodal reasoning, agents, and image and video generation.
- First model
- 2026
- Models
- 12
- Latest dated version
- Muse Spark Open
01 / RELEASE SEQUENCE
Family release sequence
- 2026-04-08Muse SparkLanguage · Reasoning
- 2026-07-07Muse ImageImage generation · Agent
- 2026-07-07Muse Video PreviewVideo generation · Multimodal
- 2026-07-09Muse Spark 1.1Reasoning · Coding
- 2026-08-05Muse Spark 1.2Reasoning · Coding
- 2026-08-10Muse Glimmer 30BLanguage · Coding
- 2026-09-01Muse Voice TranscribeLanguage
- 2026-09-02Muse Spark 1.3Reasoning · Coding
- 2026-09-23Muse Realtime AvatarVideo generation · Multimodal
- 2026-10-12Muse WonderLanguage · Reasoning
- 2026-10-12Muse Wonder FullLanguage · Reasoning
- 2026-10-15Muse Spark OpenLanguage · Reasoning
02 / GENERATIONS
What changed
Muse Image
An agentic image model that uses search and code, self-refines, and supports single- and multi-image references.
Muse Video Preview
A preview Muse video model emphasizing fidelity, temporal consistency, and native audio; not yet generally available.
Muse Spark 1.1
Improved tools, computer use, coding, and multimodal understanding with a 1M-context Meta Model API preview.
Muse Spark 1.2
The coding-focused predecessor to Muse Spark 1.3, released alongside the multi-agent Muse Code tool.
Muse Glimmer 30B
A 30B open model for always-on local agents, designed to run on one consumer GPU or a Mac or PC.
Muse Voice Transcribe
It extends Muse beyond multimodal reasoning into real-time voice input, supporting audio longer than an hour and more than 20 speakers.
Muse Spark 1.3
Compared with 1.2, Meta engineers report roughly 20% fewer tool calls and 25% fewer tokens, with stronger practical performance on long-horizon agents.
Muse Realtime Avatar
Meta describes it as an audio-driven Diffusion Transformer; it reports 448×768 portrait video at 25 fps and about 870 ms from the end of a user turn to the first byte of the synchronized voice-and-video response.
Muse Wonder
Meta introduced Muse Wonder on 2026-10-12, offering it in preview through Meta Model API and in Muse Code. Official materials do not disclose parameter count, open weights, or license.
Muse Wonder Full
Meta released Muse Wonder Full on 2026-10-12 for Muse Code subscribers. Official materials do not disclose parameter count, open weights, or license.
Muse Spark Open
Muse Spark Open extends Meta’s Muse line with downloadable full-size MoE checkpoints and local deployment and fine-tuning paths; no direct weight inheritance from Muse Spark 1.3 is inferred.