Home / Guides / Batch embedding cost savings
Batch embedding cost savings
How Batch API discounts apply to embedding jobs, when standard rates still win, and how to model savings with editable rates instead of guessed live prices.
What batch embedding cost means
Providers that offer a Batch API (or similar async bulk endpoint) often price eligible jobs lower per million tokens than the synchronous API. Embedding a large corpus is a natural fit: submit requests, wait for completion, then load vectors into the database.
Savings are real when the provider lists a Batch price for your embedding model, your job can tolerate hours of latency, and the discount applies to the token volume you send. They disappear for hot-path query embeds or when embeddings are not discounted under Batch. Always paste both rates from the provider pricing page. Rates change.
Standard vs Batch: when to use which
Prefer Batch for initial corpus index and scheduled full reindex when offered. Prefer standard for per-query question embedding and interactive tools. Tiny incremental jobs may not be worth Batch ops overhead.
Even with Batch discounts, nightly full rebuilds burn tokens. Prefer incremental updates; use Batch for the refreshes you still need.
How to calculate batch savings
Count corpus tokens honestly. Copy standard $/1M and Batch $/1M from the provider pricing page for your SKU. Standard cost ≈ tokens/1M × standard $/1M. Batch cost ≈ tokens/1M × Batch $/1M. Savings ≈ standard − Batch for that pass, then multiply by full passes per month.
The embedding cost calculator is built for this comparison: toggle Batch and keep rates editable. Batch does not remove vector DB storage, cutover engineering, or the need to re-embed when you change model or dimensions.
When Batch is the wrong lever
Query embeds at high QPS should stay on standard. Tiny corpora may save less than ops cost. Strict freshness with constant edits may favor incremental standard embeds of changed chunks over a delayed Batch full pass. If the provider has no Batch embed SKU, do not invent a discount.
Common mistakes
Avoid these Batch embedding mistakes.
- Assuming every provider’s Batch discount matches an old OpenAI marketing figure
- Sending latency-sensitive query embeds through Batch
- Counting Batch savings while still doing unnecessary nightly full reindexes
- Forgetting overlap tokens when estimating Batch job size
- Mixing Batch embed savings into RAG cost cuts without measuring generation
Frequently asked questions
- Does the Batch API reduce embedding cost?
- Often yes when the provider offers Batch pricing for your embedding model and the job is eligible. Confirm the live $/1M pair on the pricing page and model both in /tools/embedding-cost.
- How is embedding API pricing calculated with Batch?
- Same token formula, different $/1M if Batch applies: tokens/1M × Batch rate for eligible offline jobs.
- How much does text-embedding-3-small cost on Batch?
- Whatever OpenAI currently publishes for Batch vs standard for that model. Paste both; do not use evergreen articles as a price list.
- Should query embeddings use Batch?
- Almost never. Users or RAG retrieval need fast vectors. Keep queries on standard.
- How often should I re-embed if Batch is cheap?
- As rarely as quality allows. Cheap per token is not free at 10M+ tokens nightly.
- Is re-embedding the whole store every night insane?
- Often yes, even with Batch. Prefer incremental updates; Batch the full passes you still schedule.
- Do Batch discounts apply to Voyage, Cohere, or Gemini?
- Only if those providers document a bulk/async discount for embeddings. Check each pricing page; do not assume parity.
- Do embeddings charge for output tokens on Batch?
- Still input-token oriented for typical embedding jobs. Verify provider docs.
- Vector DB storage vs Batch embed API: which dominates?
- Independent. Batch helps the API line; storage follows dimensions, replicas, and corpus size.
- Where do I compare standard vs Batch quickly?
- /tools/embedding-cost with both rates pasted from the provider pricing page.
Try it in TokenCalculator
Enter corpus tokens, paste standard and Batch $/1M from your provider pricing page, and compare one-time index cost with the batch toggle on.
Open embedding cost calculator · Open cost calculator · Estimate API spend · All guides
Related tools
Related guides
Embedding cost calculator: estimate corpus spend · OpenAI embedding pricing: text-embedding-3 planning · Estimate embedding tokens · Embedding reindex cost: how often to re-embed · Embedding vs generation cost in RAG
Sources and references
Official documentation used for definitions, counting methods, or rate cards. Always confirm critical budgets on the provider page.