Most AI tasks no longer need a frontier model Read launch post

SigLIP so400m

An open image and text model that tags product photos and finds them from a shopper's words. google/siglip-so400m-patch14-384

  • Accuracy

    0.900

    Flickr30k image to text, nDCG@10

  • Time per request

    Not timed one request at a time

  • Serverless price

    $0.031

    per 1M tokens

    $0.0233

    per 1,000 images

Call it

from urllib.request import urlopen
from sie_sdk import SIEClient
​
client = SIEClient(api_key="API keysk-sie-…")
​
query = "queryA blue lace-up shoe with a white sole."
images = [
"image 1cat-01.jpg", "image 2cat-02.jpg", "image 3cat-03.jpg", "image 4cat-04.jpg",
"image 5cat-05.jpg", "image 6cat-06.jpg", "image 7cat-07.jpg", "image 8cat-08.jpg",
"image 9cat-09.jpg", "image 10cat-10.jpg", "image 11cat-11.jpg", "image 12cat-12.jpg",
"image 13arc-01.jpg", "image 14arc-02.jpg", "image 15arc-03.jpg", "image 16arc-04.jpg",
"image 17arc-05.jpg", "image 18arc-06.jpg", "image 19arc-07.jpg", "image 20arc-08.jpg",
]
model = "modelgoogle/siglip-so400m-patch14-384"
​
# Index once; keep these vectors for later searches.
items = [{"images": [{"data": urlopen("https://superlinked.com/reference/image-search/pilot/" + name).read(), "format": "jpeg"}]} for name in images]
index = []
for start in range(0, len(items), 10):
index.extend(client.encode(model, items[start:start + 10], is_query=False))
​
q = client.encode(model, [{"text": query}], is_query=True)[0]["dense"]
print(q)
print(index)
import { SIEClient } from '@superlinked/sie-sdk';

const client = new SIEClient(undefined, { apiKey: 'sk-sie-…' });

const query = "A blue lace-up shoe with a white sole.";
const images = [
  "cat-01.jpg",
  "cat-02.jpg",
  "cat-03.jpg",
  "cat-04.jpg",
  "cat-05.jpg",
  "cat-06.jpg",
  "cat-07.jpg",
  "cat-08.jpg",
  "cat-09.jpg",
  "cat-10.jpg",
  "cat-11.jpg",
  "cat-12.jpg",
  "arc-01.jpg",
  "arc-02.jpg",
  "arc-03.jpg",
  "arc-04.jpg",
  "arc-05.jpg",
  "arc-06.jpg",
  "arc-07.jpg",
  "arc-08.jpg"
];
const model = "google/siglip-so400m-patch14-384";

// Index once; preserve the JPEG bytes.
const items = await Promise.all(images.map(async (name) => {
  const response = await fetch("https://superlinked.com/reference/image-search/pilot/" + name);
  if (!response.ok) throw new Error('Could not load image');
  return { images: [new Uint8Array(await response.arrayBuffer())] };
}));
const index: { dense?: Float32Array }[] = [];
for (let start = 0; start < items.length; start += 10) {
  index.push(...await client.encode(model, items.slice(start, start + 10), { isQuery: false }));
}
const [q] = await client.encode(model, [{ text: query }], { isQuery: true });
console.log(q, index);
Build the "Image search" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: google/siglip-so400m-patch14-384 (SIE primitive: /encode). Keep the model id configurable.

Task
- Input: a text query plus one or more candidate images.
- Behaviour: return native dense vectors for the query and candidate images
- Fetch the original JPEG bytes from https://superlinked.com/reference/image-search/pilot/. Encode images in batches of 10 with is_query false, then encode the query separately with is_query true. Python: client.encode with is_query. TypeScript: client.encode with isQuery. Return the native dense vectors without application scoring or ranking.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.

Next to the other image and text models

Next to the other image and text models
Model Flickr30k image to text, nDCG@10 Time per request Serverless price Size
CLIP ViT H 14 laion2B s32B b79K laion 0.862 Not timed Run in your cloud 986M
marqo fashionSigLIP Marqo 0.838 Not timed Run in your cloud 203M
siglip so400m patch14 224 google 0.838 Not timed Run in your cloud 877M
siglip2 base patch16 224 google 0.816 Not timed $0.0231 per 1M tokens $0.0143 per 1,000 images 375M
All 10 image and text models

Measured on public datasets

Measured on public datasets
Dataset nDCG@10
Flickr30kI2TRetrieval Measured internally 0.900

Facts

Size
878M parameters
License
Apache-2.0
Reads
Text (up to 64 tokens), image
Endpoint
encode
Evaluation
Accuracy
Run on SIE against the public dataset named in each figure
Prices
Serverless list prices, effective 10 Nov 2026

Run SigLIP so400m on Serverless or in your own cloud

Native Inference

Kubernetes in your own cloud, backed by our engineers.

  • Terraform modules for your cloud
  • Apache-2.0, same engine as Serverless
  • GPU workers scale to zero between jobs
  • Separate worker pools for each tenant
Agent prompt Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Google Cloud project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Azure subscription with the superlinked/sie/azure Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Alibaba Cloud account with the superlinked/sie/alicloud Terraform module. Docs: superlinked.com/docs/deployment.

SIE Local

The open-source server on your own machine.

  • Runs on your NVIDIA GPU
  • One container, no cluster
  • Process your data fully offline
  • Same SDK and model IDs as Serverless
docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:latest-cuda12-default

Talk to us

Tell us about your use case and we'll get back to you shortly.

Apply for an inference grant

Free capacity on our hosted cluster for selected projects. Tell us what you run and we reply by email.