BGE M3
by BAAI
BAAI's multilingual workhorse, served through OpenRouter.
BGE M3 embeds 100+ languages into 1024 dimensions with strong retrieval quality for its size. The cheapest way to get good vectors, and the default for bulk indexing.
More details
BGE M3 by BAAI is a fast-speed text embedding model. BGE M3 embeds 100+ languages into 1024 dimensions with strong retrieval quality for its size. The cheapest way to get good vectors, and the default for bulk indexing. It returns 1024-dimension vectors, up to 8,192 tokens per input, batches of 100 inputs per request, multilingual coverage.
BGE M3 is priced at $0.01 per million input tokens. A million tokens is roughly 750,000 words, so a corpus of a few thousand pages costs $0.01 to embed once. You can embed text with BGE M3 on idapt.app alongside 200+ AI models in one workspace.
Pricing
Per 1M input tokens
$0.01
10k texts × 500 tokens
$0.05
Billed on input tokens only - output is vectors
Capabilities
Native dimensions
1024
Max input tokens
8,192
Batch size
100
Tuned for
any
What comes back
A vector of 1024 numbers per input — no text, no file, and nothing to play. A capture records the shape, the norm and a fingerprint so two runs can be compared without shipping the payload:
vector[1024] norm 1.000000 sha256 0f3a91c2… first 8: 0.012, -0.004, …
Illustrative shape, not an embedding of your data.
BGE M3 questions
Who makes BGE M3?
BGE M3 is made by BAAI and is served on idapt through the BAAI API.
How much does BGE M3 cost?
$0.01 per million input tokens. Embedding is billed on the text you send in, not on the vector you get back, so a few thousand pages typically cost well under a dollar.
How many dimensions does BGE M3 return?
1024 dimensions. The size is fixed for this model.
How much text can BGE M3 embed at once?
Each input can be up to 8,192 tokens, and one request can carry up to 100 inputs.
Which languages does BGE M3 support?
It is multilingual: text in most major languages can be embedded into the same vector space, so you can compare a query in one language against documents in another.
What is BGE M3 best used for?
It is tuned for general-purpose embedding. Use it to build a vector index for semantic search, retrieval-augmented generation, deduplication, clustering or classification.
Embedding a million tokens - roughly 750,000 words - costs about $0.01 on this model.