A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

BGE-M3

Compare meaning, explore text vectors, and build semantic search on your device.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Find text with a similar meaning across languages for search, clustering, and document matching. Converts each input into a 1024-number dense vector; this worker does not return sparse or multi-vector embeddings.

License
MIT
Model released
Unknown
Source repository created
Jan 27, 2024
Model weights · FP16
1.1 GB

Best for

  • Multilingual and cross-lingual dense semantic search and retrieval.
  • Clustering, similarity, and classification prototypes.

Keep in mind

  • Retrieval quality depends on language, domain, and input quality.
  • This adapter exposes dense embeddings only; BGE-M3 lexical weights and multi-vector interaction outputs are not exposed.
  • Semantic similarity is not factual verification and can encode training-data bias.
  • Long inputs are truncated and should be chunked.
  • Quantization can change ranking margins.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
BAAI/bge-m3
Release verification
Unknown
Release evidence
Exact original checkpoint public release date has not been verified. Repository creation, training, paper, artifact conversion and update dates are not release evidence.
Source updated
Jul 3, 2024
Browser model files
Xenova/bge-m3
Files updated
Feb 10, 2026
Artifact revision
4de13258303883538bd53b696b452bf8099f0858

Input

TEXT → VECTORS

Sample result

BGE-M3 — Multilingual Embeddings
Advanced settings 1
Scales each embedding vector to unit length. On makes vector lengths comparable for cosine similarity. Off preserves raw magnitudes only if supported; the listed fixed value must be retained.

Output

Explore what your text means

Generate vectors from one or more texts. Add multiple texts to compare their meaning.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

BGE-M3 — Multilingual Embeddings

16 configurations tested
Passed10
Failed6
TestedSep 10, 2026
RuntimeTransformers.js 4.2.0

Quality criteria

Finite normalized 1024d vectors, distinct rows, and positive cosine exceeds negative by at least 0.1.

PrecisionDeviceResult
FP16GPU · WebGPUAcceleratedPassedFast
Q4F16GPU · WebGPUAcceleratedPassedModerate
Q8GPU · WebGPUAcceleratedPassedModerate
UINT8GPU · WebGPUAcceleratedPassedModerate
INT8GPU · WebGPUAcceleratedPassedModerate
INT8CPU · WASMPassedModerate
Q8CPU · WASMPassedModerate
UINT8CPU · WASMPassedModerate
BNB4CPU · WASMPassedModerate
Q4CPU · WASMPassedModerate
Q4F16CPU · WASMFailed—
Q4GPU · WebGPUFailed—
BNB4GPU · WebGPUFailed—
FP16CPU · WASMFailed—
FP32CPU · WASMFailed—
FP32GPU · WebGPUFailed—
Test setup & sample details
Quality gate
Finite normalized 1024d vectors, distinct rows, and positive cosine exceeds negative by at least 0.1.
Tested
2026-09-10T10:58:00.000Z
Environment
Not recorded
Runtime
Transformers.js 4.2.0
Artifact revision
4de13258303883538bd53b696b452bf8099f0858
Fixture
{
  "file": "bge-m3.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/bge-m3.json",
  "size": 500,
  "sha1": "75a7770d06fc312ddeb39987712a8b4bbe6f1e08"
}
Rerun when
artifact revision change; runtime version change; fixture or quality gate change; browser/backend change
Per-case quality & failure details
[
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 66888.29999995232,
      "inferenceMs": 668.8000000715256,
      "totalMs": 68783.10000002384,
      "embeddingsPerSecond": 4.485645932534632,
      "grade": "slow"
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 68687.79999995232,
      "inferenceMs": 774.3999999761581,
      "totalMs": 70737.19999992847,
      "embeddingsPerSecond": 3.87396694226803,
      "grade": "slow"
    }
  },
  {
    "precision": "int8",
    "device": "wasm",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 68673.5,
      "inferenceMs": 667.2000000476837,
      "totalMs": 70675.60000002384,
      "embeddingsPerSecond": 4.49640287737649,
      "grade": "slow"
    }
  },
  {
    "precision": "int8",
    "device": "webgpu",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 67158.39999997616,
      "inferenceMs": 841.2000000476837,
      "totalMs": 69326.39999997616,
      "embeddingsPerSecond": 3.5663338086423493,
      "grade": "slow"
    }
  },
  {
    "precision": "uint8",
    "device": "wasm",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 70235.89999997616,
      "inferenceMs": 679.4000000953674,
      "totalMs": 72221.09999990463,
      "embeddingsPerSecond": 4.415660876624801,
      "grade": "slow"
    }
  },
  {
    "precision": "uint8",
    "device": "webgpu",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 90571.5,
      "inferenceMs": 775.6000000238419,
      "totalMs": 92608.80000007153,
      "embeddingsPerSecond": 3.8679731819337033,
      "grade": "slow"
    }
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "ONNX session graph error"
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 121106.29999995232,
      "inferenceMs": 513,
      "totalMs": 122968,
      "embeddingsPerSecond": 5.847953216374269,
      "grade": "slow"
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 249183.79999995232,
      "inferenceMs": 996.6000000238419,
      "totalMs": 251455.89999997616,
      "embeddingsPerSecond": 3.010234798242254,
      "grade": "slow"
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "WebGPU table index out of bounds"
  },
  {
    "precision": "bnb4",
    "device": "wasm",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 133445.09999990463,
      "inferenceMs": 801.4000000953674,
      "totalMs": 135454,
      "embeddingsPerSecond": 3.743448963866979,
      "grade": "slow"
    }
  },
  {
    "precision": "bnb4",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "WebGPU table index out of bounds"
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "ONNX session graph error"
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": 123899.80000007153,
      "inferenceMs": 247.79999995231628,
      "totalMs": 125337.10000002384,
      "embeddingsPerSecond": 12.106537532595985,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "bounded browser timeout"
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "bounded browser timeout"
  }
]

Something not quite right?

Tell us what happened with BGE-M3. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.