MetaEncoder Models by AI at Meta
AI at Meta releases · 8 in the last 12 months · this family litChangelog →
1 model2026
Last refreshed 2026-10-09. Next refresh: weekly.
Details
ResearcherAI at Meta
LicenseApache 2.0OSI-approved
Commercial useCommercial use: permitted
Models1
Released2026
Capabilities
VisionAll models
MultimodalAll models
About
MetaEncoder is Meta's family of multimodal decision-making encoders. The first model, released with an arXiv paper on 8 October 2026, turns the Muse Glimmer 30B decoder into a bi-encoder that writes both a request and each candidate option in natural language, with optional image and video inputs, and scores how well they match. Meta evaluates it on small closed-set choices and on retrieval over millions of candidates. Not a chat model family.
Decision facts
- Best fit
- embeddingvision and multimodal workcoding
- Capability starting point
- MetaEncoder with multimodal inputs
- Lowest tracked input
- Not tracked
- Closest related family
- Agents-A1
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
MetaEncoderCurrent
Use when the workload needs embedding, 29.8B parameters, and multimodal inputs.
2026-10embedding29.8B parametersmultimodal inputs
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| MetaEncoder | Use when the workload needs embedding, 29.8B parameters, and multimodal inputs. | 2026-10 | embedding29.8B parametersmultimodal inputs | Current |
Release Timeline
1 release group2026-10
1 current
MetaEncoder
Currentembedding29.8B parametersmultimodal inputs
Specifications(1 models)
| Model | Released | Parameters | Vision | Multimodal |
|---|---|---|---|---|
| MetaEncoder | 2026-10 | 29.8B | Yes | Yes |
