tomoul
ModelsPricingTrustCLIDocs
Log inStart building
BAAIflagshipoperational

bge-m3

Multilingual, multi-functionality, multi-granularity. Dense, sparse, and ColBERT outputs from a single forward pass — the workhorse for retrieval over 100+ languages.

View on Hugging Face ↗Get an API key
OverviewAPIPlaygroundBenchmarksDeploy

Quickstart

# Embed a single string
curl https://api.tomoul.ai/v1/embeddings \
  -H "Authorization: Bearer $TOMOUL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tomoul/bge-m3",
    "input": "The quick brown fox",
    "encoding_format": "float"
  }'

Specs

1024
dense dim
8,192
context
100+
languages
560M
parameters

Try it

Embedding playground

eu-helsinki1
Habari za asubuhi. Karibu Tomoul.
[ 0] -0.0142 0.0314 -0.0271 0.0083 0.0455 …
[ 1] 0.0271 -0.0145 0.0388 0.0017 -0.0223 …
[ 2] -0.0033 0.0198 0.0066 -0.0455 0.0301 …
[ 3] 0.0145 -0.0223 -0.0388 0.0271 -0.0017 …
… 1020 more dimensions
5 tokens · 178ms · $0.0000000600

Benchmarks

MTEB · Retrieval (nDCG@10)

bge-m3
59.8
openai-3-large
57.3
cohere-v3
55.9
e5-mistral-7b
55.4

Latency · eu-helsinki1 (p95)

batch=1
38ms
batch=8
66ms
batch=32
112ms
batch=128
348ms

Use it

Deployed as serverless inference in your account by default. Drop the OpenAI SDK base URL to https://api.tomoul.ai/v1, and you're running on Tomoul.