Google Gemini Embedding Pricing: gemini-embedding-2, gemini-embedding-001 & text-embedding-005 (October 2026)
Google's embedding prices are split between the Gemini API and Vertex AI pricing pages. This page puts them side by side, with the Batch API rates, the model shutdown dates and the storage effect of choosing a smaller output dimension.
gemini-embedding-2 is Google's current embedding model: stable on the Gemini API since 22 April 2026, and the replacement Google names for its older Gemini API embedding models. It charges per million input tokens, priced by modality, and nothing for output. Embedding one billion text tokens costs $200 at the standard rate or $100 through the Batch API. Source: Gemini API pricing, checked 16 October 2026.
gemini-embedding-2 Price by Input Type
| Input | Standard /M tokens | Batch /M tokens |
|---|---|---|
| Text | $0.20 | $0.10 |
| Image | $0.45 | $0.225 |
| Audio | $6.50 | $3.25 |
| Video | $12.00 | $6.00 |
Gemini API paid tier. The free tier is free of charge for every input type, but Google uses free-tier content to improve its products and the paid tier does not. Vertex AI lists the same per-million rates for Gemini Embedding 2, labelled Preview there.
All Google Embedding Models
| Model | Text $/M | Batch $/M | Status |
|---|---|---|---|
gemini-embedding-2 Text, image, video, audio, PDF 128-3072 dims, 8,192 tokens | $0.20 | $0.10 | Stable (Gemini API) |
gemini-embedding-001 Text 128-3072 dims, 2,048 tokens | $0.15 (Vertex AI) | $0.12 (Vertex AI) | Shutdown 14 May 2028 |
text-embedding-005 English text, code up to 768 dims, 2,048 tokens | Vertex meter* | Vertex meter* | Vertex AI only |
gemini-embedding-001 is no longer on the Gemini API pricing page; its rates are Vertex AI's Gemini Embedding line ($0.00015 per 1,000 online, $0.00012 per 1,000 batch). *text-embedding-005 is billed under Vertex AI's separate "Embeddings for Text (excluding Gemini Embedding)" line; see the Vertex AI pricing page for its current rate and unit. Model specs from Google's model pages; shutdown dates from the deprecations page.
Moving Off the Preview ID and gemini-embedding-001
The gemini-embedding-2 preview model was shut down on 10 August 2026; calls to the preview ID should use gemini-embedding-2. gemini-embedding-001 keeps running until its scheduled shutdown on 14 May 2028. Because the two models produce different vector spaces, switching means re-embedding the whole corpus, not just new documents. Through the Batch API that is $0.10 per million text tokens, so a corpus of 500 million tokens costs $50 to re-embed. Size your own corpus with the embedding cost calculator.
Output Dimensions: Storage Cost Impact
Both gemini-embedding-2 and gemini-embedding-001 accept an output size from 128 to 3,072 dimensions; Google recommends 768, 1,536 or 3,072. The token price is the same whatever size you request, so the saving is in downstream storage. Raw float32 size per 1 million vectors:
| Dimensions | Bytes | GiB / 1M | Ratio |
|---|---|---|---|
| 3,072 | 12,288 | 11.44 | 4x |
| 1,536 | 6,144 | 5.72 | 2x |
| 768 | 3,072 | 2.86 | 1x |
768 dimensions take a quarter of the space of 3,072. How much retrieval quality that costs depends on your data: test both sizes on a sample of real queries before committing. Price the storage itself on the vector storage page.
Vertex AI vs Gemini API: Which to Use
- - Free tier for gemini-embedding-2, rate-limited
- - Free-tier content is used to improve Google's products; paid tier is not
- - Simple per-token billing, quick setup
- - gemini-embedding-2 only on the pricing page
- - $300 Google Cloud welcome credit over 90 days for new customers
- - Prices gemini-embedding-001 and text-embedding-005 as well
- - Google Cloud billing, IAM and project quotas
- - Suits workloads already running on GCP
How gemini-embedding-2 Compares on Price
At $0.20 per million text tokens, gemini-embedding-2 costs 10x OpenAI text-embedding-3-small at $0.02/M and more than Voyage voyage-4-large at $0.12/M. What the premium buys is one vector space for text, images, video, audio and PDFs, and an 8,192-token input limit, four times gemini-embedding-001's 2,048. For text-only retrieval, compare models on your own data first; see the full embedding model comparison.