Best Embedding Models
Embedding model choice is stickier than chat model choice: switching means re-embedding your corpus, so the pragmatic questions are price per million tokens, dimensionality, and whether the model will remain available. This page compares the catalog’s embedding models on exactly those grounds, ordered by price where one is published.
An honesty note: no independent embedding quality benchmark (such as MTEB) is currently ingested into the catalog, so this page deliberately makes no quality ranking claim. It is a specification and price comparison, which for embeddings is most of the decision anyway once a model family has acceptable retrieval quality for your domain.
The top pick
The strongest measured fit for this task right now, taken from the live ranking below.
The ranking
Ordered from live catalog data; the note beneath the table states exactly how. Missing values are shown as missing, never estimated.
| # | Model | Context |
|---|---|---|
| 1 | google | 2K |
| 2 | baai | 512 |
| 3 | baai | 8K |
| 4 | baai | 512 |
| 5 | baai | 8K |
| 6 | nvidia | 10K |
| 7 | nvidia | 262K |
| 8 | qwen | 33K |
| 9 | qwen | 33K |
| 10 | google | 8K |
This list compares the 74 embedding models in the catalog on specifications and price, ordered by the lowest available price where one is published. No public quality benchmark suite for embeddings is currently ingested, so this page makes no quality claim about ordering.
marks models served through the Inferbase API. Missing values are shown as No data rather than estimated.
Also worth evaluating
The next ranks after the top ten, for teams that want a wider shortlist.
Or stop choosing manually
The honest answer changes as benchmarks and prices move. Routing makes the choice per request instead of once.
Send model="auto" to one OpenAI-compatible endpoint and each request is served by the best fit from the same quality and price data behind this ranking, with the decision disclosed per request.
Every routed request records the task it was classified as, the models that qualified, the one that won and why, so a pick made here can be checked against what the router chose in production. A model pinned by name is served as named, with no routing decision.
from openai import OpenAI
client = OpenAI(
base_url="https://api.inferbase.ai/v1",
api_key="YOUR_INFERBASE_KEY",
)
# Let routing pick the best model per request,
# or pin any model id from GET /models.
response = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": "..."}],
)Frequently asked questions
How this ranking works and how to act on it.
Because no independent embedding benchmark suite is ingested into the catalog yet, and this site does not publish rankings it cannot back with data. When embedding quality evidence is ingested, ordering will incorporate it; until then the comparison is specs and price.
Retrieval quality on your domain, price per million tokens (embedding costs scale with corpus size, not just queries), dimensionality (storage and search cost in your vector database), and operational stability, since a model change forces re-embedding.
Yes. Inferbase serves an OpenAI-compatible POST /embeddings endpoint for the served models on this page, with a pinned model id. Batch inputs preserve order, and usage is metered the same way as chat.
Embeddings from different models are not comparable: a vector index built with one model cannot be queried with another. Routing between embedding models would silently corrupt retrieval, so the endpoint requires an explicit model choice by design.
Embedding models are typically one to two orders of magnitude cheaper per token than chat models, but they process your entire corpus. For large document bases, the embedding line item is dominated by corpus size and re-embedding frequency, not query volume.
A model that turns text into a fixed-length vector so that similar meanings land near each other, the building block of semantic search, retrieval-augmented generation and clustering. Unlike a chat model it does not generate text; you call it on your documents once to index them and on each query to find the closest matches.
The smallest that keeps retrieval quality on your data. Dimension sets the storage and the search cost of every vector you keep, so a 1,024-dimension model costs about a third of a 3,072-dimension one per document in the index. Some models let you truncate the vector; the model page lists the published dimension where the creator states it.
Yes. The open-weight embedding models in the catalog can be hosted yourself or called through a provider, and the gateway serves the ones marked as served through the same embeddings endpoint. Because switching models means re-embedding the corpus, choose on price, dimension and availability first, then confirm retrieval quality on a sample of your own documents.
Related rankings
The same evidence-first method, applied to other tasks.
Put intelligence in the middle.
One gateway in front of every model, with your policies applied and every decision on record. Start on the free plan with 5,000 requests a month, no card required.