1.  [](/)
2.  [Models](/models)
3.  SigLIP 2 giant-opt patch16 (256/384)

Runs as a custom pluginEmbeddingVerified 2026-11-05

# SigLIP 2 giant-opt patch16 (256/384)

google/siglip2-giant-opt-patch16-384

SigLIP 2 giant-opt patch16 (256/384) is a image and text embedding model (Apache-2.0). Can run on Mixpeek as a custom plugin.

Mixpeek does not run it today. It can run as a custom plugin: the weights load without remote code.

[Record as JSON](/models/siglip2-giant-opt-patch16-384.json)

No demo for SigLIP 2 giant-opt patch16 (256/384) yet

A video, image, slideshow or playground goes here once we record one on our own data.

## What it is good and bad at

Best at

Best SigLIP 2 accuracy for offline catalog indexing.

Worst at

1.9B parameters and 1536-d vectors: heavy on GPU memory and index size; impractical on CPU.

Task

Embedding

Input

Image, Text

Output

Vector

Use cases

Visual search, Product search

## Templates that use it

No template uses this model yet. [Browse templates](/templates).

## Input and output

### Input

Image

Resolution

384x384 (256 sibling)

Color

RGB

Normalize

mean 0.5, std 0.5

Text

Max tokens

64

Prompts

lowercase captions; the card uses 'this is a photo of {label}.'; Gemma tokenizer, padded to max\_length 64

### Output

Vector

Dims

1536

Distance

cosine

Type

dense

Note

image and text vectors share one space; L2-normalize before cosine

## How Mixpeek wires it

Extractors

None today

Feature URIs

None today

Vector indexes

None

Retriever stages

feature\_search, mmr

Combines with

*   video frames (VideoFrameExpander) -> image vectors -> feature\_search with a text query
*   detector boxes (RT-DETR, D-FINE) -> crops -> image vectors per object
*   image vectors + E5 text vectors on captions -> feature\_search fusion (dbsf)

## How it runs

Status

Runs as a custom plugin Not served. Would run as a custom extractor plugin, loading safetensors with trust\_remote\_code false.

Serve apps

None

Runtime

transformers AutoModel + AutoProcessor (Siglip2Model)

Weights

safetensors, no remote code

Parameters

1,872 M

Size on disk

7,488 MB

Preprocessing

Resize to 384x384, rescale to 0..1, normalize mean 0.5 std 0.5; text padded to 64 tokens (max\_length).

Postprocessing

L2-normalize pooled image and text embeddings; sigmoid of scaled logits for zero-shot scores.

Serving

both, batch 32

Hardware

Runs on CPU (slower without a GPU)

### Scale to zero

Minimum replicas

0

Maximum replicas

1

Idle timeout

600 s

Memory

12 GB

CPUs

2

GPUs

1

Cold start, node warm

Not recorded

Cold start, new node

Not recorded

## Measured cost

Unmeasured

No committed measurement; plan row 3 measures it in the standalone image.

## Our evaluations

Unmeasured

We have not evaluated this model on our own data yet. Publisher results are in the sources below.

## Questions

What is SigLIP 2 giant-opt patch16 (256/384) best at?

Best SigLIP 2 accuracy for offline catalog indexing.

What are the limits of SigLIP 2 giant-opt patch16 (256/384)?

1.9B parameters and 1536-d vectors: heavy on GPU memory and index size; impractical on CPU.

Can SigLIP 2 giant-opt patch16 (256/384) be used commercially?

Yes. Apache-2.0 weights and code; no use restrictions.

How does SigLIP 2 giant-opt patch16 (256/384) run on Mixpeek?

Not served. Would run as a custom extractor plugin, loading safetensors with trust\_remote\_code false.

What does SigLIP 2 giant-opt patch16 (256/384) output?

A 1536-dimensional dense vector, compared by cosine similarity.

## Identity

Catalog id

`siglip2-giant-opt-patch16-384`

Family

siglip2

Hub id

`google/siglip2-giant-opt-patch16-384`

Variants

google/siglip2-giant-opt-patch16-256

License

Apache-2.0

Commercial use

AllowedApache-2.0 weights and code; no use restrictions.

Released

2025-02-17

Last verified

Verified 2026-11-05

Paper

[arxiv.org/abs/2502.14786](https://arxiv.org/abs/2502.14786)

Code

[github.com/google-research/big\_vision](https://github.com/google-research/big_vision)

## References and formats

*   `hf://google/siglip2-giant-opt-patch16-384@a713301b217d38485fb2204c808367d10bc3cc40`

Format

File

Loads today

safetensors

2 files, e.g. model-00001-of-00002.safetensors

Yes

## Related models

*   [SigLIP 2 so400m patch14 (224/384)](/models/siglip2-so400m-patch14-384)Embedding
*   [SigLIP 2 Base patch16 (224/256/384/512, patch32-256)](/models/siglip2-base-patch16-224)Embedding
*   [SigLIP 2 Base patch16 NaFlex](/models/siglip2-base-patch16-naflex)Embedding
*   [SigLIP 2 Large patch16 (256/384/512)](/models/siglip2-large-patch16-384)Embedding
*   [SigLIP 2 so400m patch16 (256/384/512)](/models/siglip2-so400m-patch16-384)Embedding
*   [SigLIP 2 so400m patch16 NaFlex](/models/siglip2-so400m-patch16-naflex)Embedding

## Sources

*   [huggingface.co/google/siglip2-giant-opt-patch16-384](https://huggingface.co/google/siglip2-giant-opt-patch16-384)
*   [arxiv.org/abs/2502.14786](https://arxiv.org/abs/2502.14786)
*   [huggingface.co/docs/transformers/model\_doc/siglip2](https://huggingface.co/docs/transformers/model_doc/siglip2)