APIRank
EN 中文

Chroma

Listed at https://www.trychroma.com

Overall Rank #22 ⭐ Consider
✅ Chroma OSS runs fully local (on any server, zero latency); Chroma Cloud on AWS us-east-1 / eu-west-1 / ap-southeast-2 (Singapore node planned H2 2026), ~200-400ms latency from mainland China | 🌍 International

💰 Token Pricing

TypePriceNote
Input Chroma OSS 完全免费(Apache 2.0),本地/嵌入式运行;Chroma Cloud Free 永久层:50K 向量、5K 查询/月、0.5 GB 存储;Pro $0.30/M 向量-月 + $0.01/M 查询(月最低 $10);Enterprise 合同制(自定义节点、HIPAA、SOC 2) per million tokens
Output 存储(GB-月)+ 向量数 + 查询数 三维度计费;无 per-embedding API 费;Default Embedding Function 本地推理零成本;BYO Embedding 由第三方 API 计费 per million tokens
💡 Free Credits: Chroma OSS permanently free, Apache 2.0, no feature limits, no vector count cap (limited by local hardware); Chroma Cloud Free forever: 50K vectors + 5K queries/mo + 0.5 GB storage + 1 Collection + community support

🤖 Supported Models (12)

All-MiniLM-L6-v2 (default, 384-dim, SBERT)All-mpnet-base-v2 (768-dim, higher quality SBERT)multi-qa-MiniLM-L6-cos-v1 (768-dim, QA-tuned)BGE-small-en-v1.5 (384-dim, BAAI)BGE-base-en-v1.5 (768-dim, BAAI)BGE-large-en-v1.5 (1024-dim, BAAI)E5-base-v2 / E5-large-v2 (intfloat, 768/1024-dim)Cohere embed-english-v3.0 / embed-multilingual-v3.0 (via API)OpenAI text-embedding-3-small / text-embedding-3-large (via API)VoyageAI voyage-3 / voyage-large-2 (via API)Hugging Face transformers2vec (any Sentence-Transformers / ONNX model)InstructorXL / BGE-M3 (instruction-tuned, multilingual)

✨ Pros

  • ✓🦄 21k+ GitHub stars (LangChain/LlamaIndex default vector store), Python ecosystem de-facto standard
  • ✓Apache 2.0 fully open-source, supports embedded (pip install chromadb), client-server, Cloud three deployment modes
  • ✓Python-first API: collection.add() / query() / get() one-liner, ideal for rapid prototyping and Jupyter experiments
  • ✓Default embedding function (Sentence-Transformers) runs locally at zero cost; supports 12+ embedding models + BYO OpenAI/Cohere/Voyage
  • ✓Built-in metadata filtering / Where clauses, $eq / $in / $gte multi-condition composition, no backend SQL needed
  • ✓DuckDB + Parquet persistence layer (embedded mode), single-machine million-vector workloads no sweat

⚠️ Cons

  • ×Cloud has few regions (only 3 in 2026 H1), no GPU index acceleration (Pinecone/Qdrant both have it)
  • ×Horizontal scaling / sharding weaker than Pinecone (>10M vectors consider migrating to Qdrant/Weaviate)
  • ×Cloud Pro starts $10/mo + usage billing, large vector set cost higher than Pinecone Serverless
  • ×No native hybrid search (Sparse+Dense, Qdrant is benchmark), need to manually compose BM25
  • ×High latency from mainland China to Cloud (OSS mode can self-host domestically)

🎯 Best For

RAG prototyping and production for Python data science teams (Jupyter / LangChain / LlamaIndex); cost-sensitive SaaS at small/medium scale (<5M vectors); local LLM apps needing embedded in-process vector retrieval; zero-cost vector DB for teaching and research scenarios