Question 1

What vector dimensions should I use for OpenAI embeddings?

Accepted Answer

text-embedding-3-small and text-embedding-3-large support configurable dimensions via Matryoshka Representation Learning (MRL). For most RAG applications, text-embedding-3-small at 1536 dims is the default and offers the best cost/performance balance. You can reduce to 512 dims for ~2.5× storage savings with minor quality loss. text-embedding-ada-002 is fixed at 1536 dims. For production with millions of documents, consider int8 quantization to cut storage costs by 75%.

Question 2

When should I use cosine similarity vs dot product vs euclidean?

Accepted Answer

Cosine similarity measures the angle between vectors (scale-invariant) — best for semantic search where you care about meaning, not vector magnitude. Use cosine for text embeddings from any provider. Dot product is equivalent to cosine when vectors are normalized (unit length), and is faster to compute — Pinecone normalizes by default. Euclidean distance measures straight-line distance in vector space — useful for image embeddings or cases where magnitude encodes information. For text search, always start with cosine.

Question 3

What is the difference between Pinecone, Weaviate, Qdrant, and ChromaDB?

Accepted Answer

Pinecone is a fully managed cloud vector database — zero infrastructure, generous free tier (1 index, 100K vectors), ideal for prototypes and production. Weaviate is open-source (self-host) or managed cloud, with built-in ML inference and GraphQL API — best for hybrid search (vector + BM25). Qdrant is a high-performance open-source vector DB with advanced filtering, payload indexing, and quantization — great for on-premises or Docker deployments. ChromaDB is an embedded Python library (no server needed) — best for local development, notebooks, and small-scale apps.

Question 4

How do I calculate storage requirements for vector indexes?

Accepted Answer

Storage per vector = dimensions × bytes_per_float. float32 = 4 bytes/dim, float16 = 2 bytes/dim, int8 = 1 byte/dim. For 1M vectors at 1536 dims float32: 1,000,000 × 1536 × 4 = ~5.86 GB raw. Add metadata/payload overhead (typically 20-30%). Plus the HNSW graph structure adds ~60 bytes per vector. Real-world 1M vector index at 1536 dims ≈ 8-12 GB total. Managed services charge based on pod type — a Pinecone p1.x1 pod holds ~1M 1536-dim vectors at ~$70/month.

Question 5

What is pgvector and when should I use it instead of a dedicated vector database?

Accepted Answer

pgvector is a PostgreSQL extension that adds vector storage and similarity search directly to your existing Postgres database. Use pgvector when: you're already running PostgreSQL and want to avoid adding another service, your vector count is under 1-2M (beyond that, dedicated vector DBs outperform), or you need ACID transactions combining vector and relational data. Limitations: slower than dedicated vector DBs at scale, no built-in quantization, limited indexing options (HNSW and IVFFlat). For new projects without an existing Postgres stack, Qdrant or Pinecone are generally better choices.

Model	Provider	Dims	Cost/1M tokens	Notes
text-embedding-3-small	OpenAI	1,536	$0.02	Best price/perf, MRL configurable (512/1024/1536)
text-embedding-3-large	OpenAI	3,072	$0.13	Highest quality, MRL configurable (256/1024/3072)
text-embedding-ada-002	OpenAI	1,536	$0.10	Legacy, fixed dimensions, widely supported
all-MiniLM-L6-v2	HuggingFace (local)	384	Free	Fast, small, great for local/offline use
all-mpnet-base-v2	HuggingFace (local)	768	Free	Higher quality than MiniLM, still fast
nomic-embed-text	Nomic / local	768	Free	Strong open-source, Ollama compatible
voyage-3	Voyage AI	1,024	$0.06	Competitive MTEB, good for RAG retrieval
voyage-3-lite	Voyage AI	512	$0.02	Budget option, smaller vectors
embed-english-v3.0	Cohere	1,024	$0.10	Multilingual available (embed-multilingual-v3)
BGE-large-en-v1.5	BAAI (local)	1,024	Free	Top open-source MTEB, CPU-friendly
jina-embeddings-v2-base	Jina AI	768	$0.02	8192 token input context, long documents

Vector Search Config Builder

Select Provider

Index Configuration

Storage Estimator

Embedding Model Dimension Reference

Related Products

Developer Productivity Bundle

AI Prompt Engineering Toolkit

Developer Cheat Sheet Bundle

You Might Also Need

JSON Formatter & Validator

Regex Tester

Base64 Encoder & Decoder

More Free Tools

Related Articles

curl 深度使用手冊：API 測試從入門到進階

Best AI Code Review Tools 2026: CodeRabbit, Sourcery, DeepCode & Bito Compared

AI Code Review Tools 2026: GitHub Copilot, CodeRabbit, Qodo Merge Ranked

AI Coding Tools Comparison 2026: GitHub Copilot vs Cursor vs Windsurf vs Cline

AI Pair Programming in 2026: GitHub Copilot vs Cursor vs Claude Code

Enjoying this tool? Get more like it.

Unlock AI-Powered Dev Tools

Get DevPlaybook Pro