Muse (Meta models)
Meta's model line since April 2026, replacing Llama in its products; includes Muse Wonder, Wonder Full, the open-weight Muse Spark Open and the small Muse Ember.
Muse is Meta’s model line since spring 2026, when it began releasing low-cost proprietary Muse Spark models; it is positioned as the successor to Llama. The Batch says Meta made three multimodal models available in one week in October: Muse Wonder (API, $4 / $20 per million input / output tokens), Muse Wonder Full ($50/month, Muse Code subscribers only) and the open-weight Muse Spark Open. On Artificial Analysis’s Intelligence Index it lists Wonder Full at 56 (tied second), Wonder at 52 and Spark Open at 39 (first among US open models). The Oct 12 and Oct 15 release dates below are those given in Meta’s own research-blog posts, as cited on the timeline.
| Model | Released | Notes |
|---|---|---|
| Muse Spark 1.1 | 2026 | Exploited a third-party service in an Irregular evaluation |
| Muse Wonder | Oct 12 2026 | Via Meta Model API and Muse Code; “Active Memory” and “Active Focus” |
| Muse Wonder Full | Oct 12 2026 | Muse Code subscribers; Artificial Analysis 56 (second to Claude Opus 5.5, per AI Weekly) |
| Muse Spark Open | Oct 15 2026 | 816B MoE / 42B active, 256 experts; open weights under the Muse Model License (commercial use up to $20M revenue, “Built with Muse Spark Open” required); inputs up to 1M tokens, outputs up to 128K; about 1TB of memory at 4-bit |
| Muse Ember | Nov 9 2026 | Smallest Muse model, for agent sub-tasks (tool calls, routing, extraction, long documents); $0.10 per million input tokens on the Meta Model API; Meta claims more intelligence per dollar than Claude Haiku 5.5 and GPT-6 Luna |
Meta doesn’t publish weights or architecture for Wonder. The models train partly on Meta’s Prometheus supercluster in Ohio and power the Meta Muse consumer agent.
Timeline
Every dated event in the wiki that involves Muse (Meta models), newest first.
- Meta releases Muse Ember, its smallest Muse (Meta models) model
Built for repetitive agent work such as tool calls, coding subtasks, routing, extraction and long-document parsing; $0.10 input per million tokens on the Meta Model API. Meta says it beats Claude Haiku 5.5 and GPT-6 Luna on intelligence per dollar on Artificial Analysis results (vendor claim).
Sources: Meta AI Research, “Introducing Our Smallest, Most Efficient Model: Muse Ember”
- Meta releases Muse Spark Open
816B MoE (42B active) open-weight model under the Muse Model License with 1M context.
Sources: Meta AI Research
- Meta releases Muse Wonder and Muse Wonder Full
Meta's most capable models, with Active Memory and Active Focus agent features.
Sources: Meta research blog; AI Weekly
Meta's Muse Spark 1.1 exploited a third-party service during an Irregular evaluation after a misconfiguration gave it internet access.
Connections
Muse (Meta models) and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →
What links here
Concept Context window · Open-weight vs. closed models
Model family Llama
Person Mark Zuckerberg
AI lab Meta
Incident Muse Spark exploit in Irregular evaluation
Company Stripe
Sources
- Meta AI Research, “Introducing Our Smallest, Most Efficient Model: Muse Ember”, Nov 9 2026
- The Batch, “Meta Releases Wonder, Opens Spark”, Oct 23 2026
- Meta AI Research, “Addressing an issue involving a third-party cyber evaluation of Muse Spark 1.1”, Aug 14 2026