Embedding Model Pricing
Gemini Embedding 2 costs $0.20 per 1M text tokens ($0.10 batch, $0.45 for images). OpenAI's text-embedding-3-small is 10x cheaper at $0.02, and text-embedding-3-large sits at $0.13. This page compares current embedding prices across Google, OpenAI, Voyage AI, and Mistral, with a calculator for your corpus.
Prices as of July 2026
Embedding prices compared
All prices are USD per 1M input tokens. Embeddings bill input only; there are no output token charges.
| Model | Provider | $/1M tokens | Batch $/1M | Input limit | Dimensions | Modalities |
|---|---|---|---|---|---|---|
| Gemini Embedding 2 | $0.20 | $0.10 | 8,192 | 128-3,072 | Text, image, audio, video, PDF | |
| Gemini Embedding 001 | $0.15 | n/a | 2,048 | 768-3,072 | Text | |
| text-embedding-3-large | OpenAI | $0.13 | n/a | 8,191 | 256-3,072 | Text |
| text-embedding-3-small | OpenAI | $0.02 | n/a | 8,191 | 512-1,536 | Text |
| text-embedding-ada-002 | OpenAI | $0.10 | n/a | 8,191 | 1,536 | Text |
| voyage-4-large | Voyage AI | $0.12 | n/a | 32,000 | 256-2,048 | Text |
| voyage-4 | Voyage AI | $0.06 | n/a | 32,000 | 256-2,048 | Text |
| voyage-4-lite | Voyage AI | $0.02 | n/a | 32,000 | 256-2,048 | Text |
| Mistral Embed | Mistral | $0.10 | $0.050 | 8,000 | 1,024 | Text |
| Codestral Embed | Mistral | $0.15 | $0.075 | 8,000 | Flexible (first-n) | Text (code) |
Gemini Embedding 2 image inputs bill at $0.45/1M tokens ($0.225 batch). Prices verified July 2026 from official provider pages: Google, OpenAI, Voyage AI, Mistral.
What will your corpus cost?
Embedding cost is a one-time indexing bill: number of documents times average tokens per document times the per-token rate. Enter your corpus size or pick a preset.
One-Time Embedding Cost Calculator
Embedding 500M tokens costs $10.00 with text-embedding-3-small vs $100.00 with Gemini Embedding 2. It is a one-time cost: you pay once per token embedded, then only for new or changed documents and query embeddings.
| Batch cost | |||
|---|---|---|---|
| text-embedding-3-smallOpenAICheapest | $0.02 | $10.00 | n/a |
| voyage-4-liteVoyage AI | $0.02 | $10.00 | n/a |
| voyage-4Voyage AI | $0.06 | $30.00 | n/a |
| text-embedding-ada-002OpenAI | $0.10 | $50.00 | n/a |
| Mistral EmbedMistral | $0.10 | $50.00 | $25.00 |
| voyage-4-largeVoyage AI | $0.12 | $60.00 | n/a |
| text-embedding-3-largeOpenAI | $0.13 | $65.00 | n/a |
| Gemini Embedding 001Google | $0.15 | $75.00 | n/a |
| Codestral EmbedMistral | $0.15 | $75.00 | $37.50 |
| Gemini Embedding 2Google | $0.20 | $100.00 | $50.00 |
How to choose an embedding model
For plain text retrieval on a budget, start with text-embedding-3-small or voyage-4-lite at $0.02 per 1M tokens. A 500M-token corpus costs about $10 to embed with either, versus $100 with Gemini Embedding 2 or $65 with text-embedding-3-large. For most RAG applications the retrieval quality gap between small and large models is narrower than the 6x to 10x price gap, so benchmark on your own data before paying the premium. If a provider offers a batch API, use it for the initial index: it halves the price and one-time indexing rarely needs real-time latency.
If your corpus is not just text, Gemini Embedding 2 is the standout: it embeds text, images, audio, video, and PDFs into one vector space, so a screenshot and the paragraph describing it land near each other in search. Text costs $0.20 per 1M tokens and images $0.45 per 1M, which is premium pricing, but the alternative is running separate text and image embedding pipelines and stitching the results together. For document-heavy workloads with figures and scanned pages, one multimodal model is usually simpler and cheaper in engineering time than two single-modal ones.
Dimensions are a second bill hiding behind the first. Vector databases charge for storage and memory, and a 3,072-dimension vector costs roughly 4x as much to store and search as a 768-dimension one, forever. Models with Matryoshka-style flexible output (Gemini Embedding 2 down to 128, text-embedding-3-large down to 256, the voyage-4 series down to 256) let you truncate vectors with modest quality loss. If you are embedding tens of millions of documents, a cheap-to-embed model at high dimensions can end up costing more per month in vector storage than an expensive model truncated to 512.
Related Tools
Frequently Asked Questions
Quick answers about embedding model pricing