Gemini Embedding vs OpenAI embeddings

Gemini Embedding

Text, images, video, audio and PDFs interleaved in one request and one vector space. $0.20 per million text tokens, against $0.02 for OpenAI's small model.

OpenAI embeddings

text-embedding-3-small at $0.02 per million tokens, $0.01 through the Batch API. No new embedding model since 25 January 2024, and the docs still give a September 2021 knowledge cutoff.

Gemini Embedding

1. Don't send task_type to gemini-embedding-2. Prefix the text instead, task: search result | query: ... for queries and title: ... | text: ... for documents
2. Ask for output_dimensionality 768 unless you need 3072. Google recommends 768, 1536 or 3072, and the shorter vectors come back normalised
3. Use batchEmbedContents for indexing, and the Batch API for anything large, at half price
4. Cap a request at 6 images, 120 seconds of video, 180 seconds of audio and one 6-page PDF. Split longer media first
5. Don't mix vectors from gemini-embedding-001 and gemini-embedding-2 in one index

OpenAI embeddings

1. Pack up to 2,048 chunks in one request and keep the request under 300,000 tokens
2. Count tokens before sending. An input over 8,192 tokens is rejected, not truncated
3. Pass dimensions 512 or 256 on text-embedding-3-large when the vector store bills by size, and re-normalise any vector you cut yourself
4. Split a Batch API index job into batches of under 50,000 inputs. It's half price with a 24-hour window
5. Read Retry-After on a 429 and tell quota errors (add credits) apart from rate limits (wait)