7 models · Embeddings · Transformers.js

Embeddings in the browser

7 models handle embeddings in the browser today, from 28.6 MB (all-MiniLM-L6-v2) to 667.5 MB (BGE-M3), a 201.3 MB download at the midpoint. Every one runs with Transformers.js over WebGPU or WebAssembly: no server, no install, no account.

Models: 7
Smallest: 28.6 MB
Median: 201.3 MB
Largest: 667.5 MB

Every embeddings model, ranked by WebGPU size

Model WebGPU · WASM

Model	WebGPU download	WASM download
all-MiniLM-L6-v2 22.7M params	28.6 MB q4f16	21.8 MB uint8
bge-small-en-v1.5 33.4M params	34.5 MB q4f16	32.2 MB uint8
EmbeddingGemma-300M 302.9M params	168.0 MB q4f16	295.1 MB q8
voyage-4-nano 346.45M params	201.3 MB q4f16	402.3 MB q8
GTE Multilingual Base 305M params	443.7 MB q4f16	324.6 MB uint8
Qwen3-Embedding-0.6B 595.78M params	541.2 MB q4f16	585.1 MB uint8
BGE-M3 569M params	667.5 MB q4f16	542.1 MB uint8

Sizes measured from the HuggingFace API file tree for each model's repo, not estimated. Sorted smallest to largest by the WebGPU headline pick.

Size ladder

The WebGPU download spans 23.3x here: 28.6 MB (all-MiniLM-L6-v2) to 667.5 MB (BGE-M3).

4 of 7 models download a smaller build over WebAssembly than WebGPU: the CPU-fallback quant compresses tighter than the GPU pick there.

2 of 7 models here fit under 100 MB over WebGPU.

Bytes-per-million-params ranges 2.6x within this group: 0.55 MB/M (EmbeddingGemma-300M) to 1.45 MB/M (GTE Multilingual Base).

Use it with Transformers.js

Every embeddings model here shares the same pipeline() task, so one snippet covers the group. This one loads all-MiniLM-L6-v2, the smallest download; swap the repo string for any other row in the table above.

import { pipeline } from "@huggingface/transformers";

const pipe = await pipeline("feature-extraction", "Xenova/all-MiniLM-L6-v2", {
  device: "webgpu", // usually falls back to "wasm"; wrap in try/catch for production
  dtype: "q4f16", // use "uint8" for the WASM build
});

const result = await pipe(/* your input */);

Requires npm install @huggingface/transformers (or the CDN build).

Notable embeddings models

Smallest: all-MiniLM-L6-v2 28.6 MB Typical size: voyage-4-nano 201.3 MB Largest: BGE-M3 667.5 MB

Other tasks

Speech to text 7 Text generation 13 Text to speech 2 Vision 12 Vision + language 5 Multimodal (any-to-any) 3 Text utilities 3

Or see every browser model grouped by task, or the main GGUF catalog for native runtimes outside the browser.

FAQ

Do all embeddings models here use the same WebGPU quant?

Yes: all 7 models here use the q4f16 WebGPU build.

What would it cost to download every embeddings model here?

2.04 GB total over WebGPU across all 7 models, if you tried every one back to back. Most projects only need the single model that fits the job, not the whole set.

Sizes measured 2026-08-01 from the HuggingFace API. Last validated 2026-09-14. See all browser models or the methodology.