Primary use stated or inferred cautiously from official documentation.
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.
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
Limits and integration
| API ID | facebook/meta-encoder-v1 |
|---|---|
| Model type | Embeddings · Vision and documents |
| Access model | Open weights |
| License | Apache-2.0 |
| Deployment | Self-hosted |
| Release | 12 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image · Video |
| Exit | Embeddings · Probabilities |
| Context window | Training sequences up to 16,384 tokens; no automatic truncation |
| maximum output | Normalized 6,656-dimensional embeddings or typed decisions |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch · Transformers |
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.
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.
What it can do
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.
Context
Training sequences up to 16,384 tokens; no automatic truncation · Normalized 6,656-dimensional embeddings or typed decisions
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous 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.
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.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| MMEB V3 image | 0.816 | Published score · open-set | View source ↗ |
| MMEB V3 video | 0.743 | Published score · open-set | View 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.
Chronology
facebook/meta-encoder-v1
Weights and model card publication.
7615b21a
Apache-2.0 license added to the repository; revision verified for the catalog.