Embeddings

Tool id	Vendor	Credentials	Status	Divinci usage
@cf/baai/bge-base-en-v1.5 transforms any given text into a 768-dimensional vector	BAAI	Divinci-managed	available	43 (0%)
@cf/google/embeddinggemma-300m multilingual Gemma embedding model producing 768-dimensional vectors	Google	Divinci-managed	available	64629 (99%)
gemini-embedding-001@1536 Google's gemini-embedding-001 model truncated to 1536 dimensions — recommended for qdrant-cosine and other 1536d collections.	Google	Divinci-managed	available	363 (1%)
gemini-embedding-001@3072 Google's gemini-embedding-001 at its full 3072 dimensions — highest retrieval quality. Not compatible with Cloudflare Vectorize (1536 max) or Couchbase (2048 max).	Google	Divinci-managed	available	37 (0%)
gemini-embedding-2-preview@1536 Google's gemini-embedding-2-preview model truncated to 1536 dimensions via Matryoshka. 8K token input, multimodal capable. Best balance for qdrant-cosine and other 1536d collections.	Google	Divinci-managed	available	297 (0%)
vertex-ai-text-embedding-005 transforms any given text into a 768-dimensional vector using Google Vertex AI	Google	Divinci-managed	available	0
voyage-3.5 Previous-generation Voyage AI general-purpose embedding model	Voyage AI	BYOK	available	0
voyage-4 Voyage AI's balanced general-purpose embedding model	Voyage AI	BYOK	available	0
voyage-4-large Voyage AI's highest-quality general-purpose embedding model	Voyage AI	BYOK	available	0
voyage-4-lite Voyage AI's low-latency, low-cost embedding model	Voyage AI	BYOK	available	1 (0%)

Usage: 65,493RAG vectors in production, counted 2026-10-16. Refreshed weekly.

Never mix embedding models inside one vector

Vectors produced by two different models live in incompatible spaces, even at identical dimensionality. Mixing them does not error — similarity search just quietly gets worse, in a way that looks like "the corpus isn't very good". The vector's embeddingModel is the source of truth; do not override it per request.

Picking by ingestion shape, not by quality alone

The two Gemini embedding models differ far more in rate limit than in output quality, and that is usually what decides:

Model	Requests/min	API mode	Use for
gemini-embedding-001@1536	~100K	batched	Catalog imports and any batch over ~100 files. Survives parallel ingestion.
gemini-embedding-2-preview@1536	~4K	unbatched	Individual uploads and quality-sensitive corpora. Saturates immediately under batch load.

Practical rule:

• A single PDF, transcript or page — either model finishes inside the window, even with hundreds of chunks.
• A catalog of 30 or fewer items — either works; the preview model will hit its cap briefly and the retry chain absorbs it.
• A bulk import, or a fresh corpus reseed — use gemini-embedding-001, or ingest through a temporary vector on it and re-embed later.

The failure signature of getting this wrong is a pile of Temporary API error (429) / Quota exceeded for … online_prediction_requests entries in the vector's failed sources. That is a rate cap, not a bug: retries do eventually succeed, at the cost of hours of wall-clock.

Voyage AI

voyage-4-large, voyage-4, voyage-4-lite and voyage-3.5 are 1024-dimension models from Voyage AI, reached through MongoDB Atlas's hosted endpoint or Voyage's own. What is different about them:

• Two ways to pay for the calls. Where Divinci has enabled its own Voyage key you can use it; you can also save your own key (provider VOYAGE, under Tools → Providers) and attach it to a vector. With your own key the embedding calls go straight to your Voyage account and Divinci adds no embedding charge on top; a step that costs Divinci compute is still billed as usual. GET /api/v1/rag/embedding-models reports, per model, whether it takes divinci's key, byok only, or either.
• Asymmetric. Documents and queries are embedded with different input_types. Divinci sets this for you: documents at ingest, queries at search.
• A lower similarity scale than Gemini or BGE. Voyage cosine scores for relevant passages sit lower, so a new Voyage vector defaults to a minimum similarity of 0.40, not 0.62 (at 0.62 only about 15–50% of relevant passages survived in our measurements on SciFact and SciDocs). That is two corpora, so tune minimumSimilarity for yours.
• Not recommended by default. On SciFact, voyage-4 scored below gemini-embedding-001 in our retrieval lab. They are offered for teams that already use Voyage, or want to compare.

Changing to or from Voyage needs a new vector

Moving a vector that already holds embeddings onto or off a Voyage model is refused: the spaces are not comparable, and a same-dimension swap would not error, it would just retrieve badly. Create a new vector with the model and upload files to it directly. Re-indexing through the queued path (crawls, connectors, migrate) does not support Voyage yet.

Dimensions

1536 and 3072 variants of gemini-embedding-001 are separate tools because dimensionality is part of the vector's contract — the store is created at that width. Larger is not free: it costs proportionally more storage and query time for a gain that is corpus-dependent.

See also

• RAG & Vector Stores
• RAG Knowledge Base