Vector Dimension Comparison
Compare vector dimensions, context length, MTEB scores and prices of mainstream embedding models, with filtering and sorting, plus storage estimation for millions of text chunks.
FreeOnline Tool
Loading…
How to Use
- Filter by dimension range: small (at most 768), medium (769-1536) or large (at least 1537).
- Filter by price: free/open-source, low (under $0.05/M), medium ($0.05-$0.20/M) or high (at least $0.20/M).
- Click any column header (model name, provider, dimension, max tokens, MTEB score, price) to sort ascending or descending.
- Rows highlighted in green are recommended models with a 'Recommended' badge.
- Enter the number of text chunks in the storage estimator; the tool computes storage for representative models per dimension assuming float32 (4 bytes per dimension).
Features
- 21 mainstream embedding models from OpenAI, Cohere, Voyage AI, BAAI, Alibaba, Microsoft, Jina AI and more.
- Each row shows vector dimension, available dimension options, max tokens, MTEB score and price per million tokens.
- Dual filters for dimension range and price range, plus click-to-sort on every column.
- Recommended models are marked with a green background and a 'Recommended' badge.
- Storage estimator: enter your chunk count and get storage usage per dimension at float32 precision (4 bytes per dimension).
- MTEB (Massive Text Embedding Benchmark) is a comprehensive English text embedding benchmark for horizontal comparison.
Use Cases
Model selection for RAG
Before building a RAG pipeline, compare dimensions, MTEB scores and prices to pick an embedding model that fits the budget.
Storage planning
Enter your planned chunk count to estimate vector storage across dimension options and decide whether to reduce dimensions.
Budget-friendly filtering
Use the price filter to quickly locate free or low-cost models with the best performance within budget.
Score-based ranking
Sort by MTEB score to find top-performing models, then balance against price and context length.
FAQ
Is the data up to date?
The table is built from static reference data for comparison purposes; the page notes that official vendor docs are authoritative. Some models (e.g. OpenAI text-embedding-3) support multiple output dimensions.
What does the MTEB score mean?
MTEB (Massive Text Embedding Benchmark) is a comprehensive benchmark for English text embeddings; a higher score means better overall performance on that benchmark.
How is storage calculated?
At float32 precision: storage = chunk count x dimensions x 4 bytes. int8 quantization or MRL dimension reduction would use less.
Why do some rows list multiple dimensions?
Some models can output vectors at several dimensions (e.g. text-embedding-3-small supports 512 or 1536); the available options are listed in the row.
What is the basis for the Recommended badge?
A manual balance of moderate dimension, good MTEB score and reasonable price, a good starting point for most scenarios. Final choice should be validated against your own data and benchmarks.