M
META / facebook/meta-encoder-v1

MetaEncoder v1

Meta multimodal encoder for representing text, images, visual documents and frame sequences in a shared space. Weights were published on November 12 and the repository added the Apache-2.0 license on November 17. It provides embeddings, retrieval and decisions among options.

01 / SUMMARY

The essential

ORIENTATIONMultimodal embeddings and retrieval

Primary use stated or inferred cautiously from official documentation.

CONTEXTTraining sequences up to 16,384 tokens; no automatic truncation

Maximum input capacity when published by the source.

EXITNormalized 6,656-dimensional embeddings or typed decisions

Documented maximum generation limit.

ACCESSHugging Face · downloadable checkpoint

Verified availability channels.

MODEL TYPEEmbeddings · Vision and documents

Primary functional family and verified specialties.

LICENSEApache-2.0

Declared terms for API access, model weights, or self-hosting.

02 / TECHNICAL SHEET

Limits and integration

API IDfacebook/meta-encoder-v1
Model typeEmbeddings · Vision and documents
Access modelOpen weights
LicenseApache-2.0
DeploymentSelf-hosted
Release12 NOV 2026
Knowledge cutoffNot published
EntranceText · Image · Video
ExitEmbeddings · Probabilities
Context windowTraining sequences up to 16,384 tokens; no automatic truncation
maximum outputNormalized 6,656-dimensional embeddings or typed decisions
ReasoningNot published
Published tools
Structured OutingsNot published
Batch processingNot published
Prompt cacheNot published
Fine-tuningNot published
Verified platformsHugging Face · PyTorch · Transformers
BEST SUITED FOR

Multimodal search, clustering and document retrieval. Query and corpus instructions may differ. The card states that one 80 GB GPU holds the BF16 weights; total memory also depends on input and batch size.

WORTH MONITORING

It is not a text generator. A training context of 16,384 tokens does not by itself establish a larger operational maximum. The wrapper controls templates and pooling; changing instructions or truncating alters the representation. Review the code before loading the model.

03 / CAPABILITIES

What it can do

01

Orientation

Multimodal search, clustering and document retrieval. Query and corpus instructions may differ. The card states that one 80 GB GPU holds the BF16 weights; total memory also depends on input and batch size.

02

Context

Training sequences up to 16,384 tokens; no automatic truncation · Normalized 6,656-dimensional embeddings or typed decisions

03

Tools and integration

Not published

04

Access

Hugging Face · downloadable checkpoint

Input modalities
TextImageVideo
04 / PRICES

Documented cost

ConceptWorthUnit/condition
Standard inputNo published official API rateStandard API rate
Cached inputNot publishedReading reused prefixes
Cache write or storageNot publishedThe condition varies by provider
Standard outputSelf-hosted infrastructure costMay include reasoning tokens
Batch inputNot publishedAsynchronous processing
Batch outputNot publishedAsynchronous processing

The Apache-2.0 license is published in the pinned revision. No official API tariff has been verified; self-hosting consumes memory and compute. The November 17 verification does not change the original weights date.

i Prices change and may depend on level, region or context length. Check the source before making a decision.

05 / EVALUATIONS

How to read the results

“

The repository table reports 0.816 on MMEB V3 image and 0.743 on MMEB V3 video under open-set evaluation. Closed-set results compare options rather than generating LLM answers. These are manufacturer figures, not tests reproduced by Inferama.

BenchmarkResultMetricSource
MMEB V3 image0.816Published score · open-setView source ↗
MMEB V3 video0.743Published score · open-setView source ↗

Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.

06 / VERSIONS

Chronology

facebook/meta-encoder-v1

Weights and model card publication.

7615b21a

Apache-2.0 license added to the repository; revision verified for the catalog.

07 / SOURCES

Traceability