I'm building a support search app and need an embedding model recommendation

For a support search app, I’d recommend using a general-purpose text embedding model that’s strong at semantic retrieval, rather than a chat or generation model.

Best default choice

OpenAI text-embedding-3-large
Use this if you want the highest retrieval quality and can afford a bit more latency/cost.

Why it’s a good fit

• Strong semantic matching for support tickets, FAQs, KB articles, and user queries
• Good at capturing paraphrases and intent
• Works well for hybrid search setups with keyword + vector retrieval
• Handles short queries and longer support content effectively

If you want lower cost / faster

OpenAI text-embedding-3-small
Choose this if:

• You have high volume
• You want cheaper indexing and query-time embeddings
• You can accept slightly lower quality than the large model

Good practical setup for support search

A strong production pattern is:

1. Chunk help docs / articles into sections
2. Embed each chunk
3. Store vectors in a vector DB
4. At query time:
   • Embed the user query
   • Retrieve top-k nearest chunks
   • Optionally rerank with a cross-encoder or LLM reranker
   • Return answer/snippets

Recommendation by use case

• Best quality: text-embedding-3-large
• Best cost/performance: text-embedding-3-small
• If you need open-source/on-prem: consider bge-large-en, e5-large, or GTE-large

My suggestion

If you’re just starting:

• Start with text-embedding-3-small for a pilot
• Switch to text-embedding-3-large if search quality isn’t good enough

If you want, I can also recommend:

• the best chunking strategy for support docs
• a vector DB option
• a full search architecture for your app