Home / Guides / Embedding reindex cost: how often to re-embed
Embedding reindex cost: how often to re-embed
When to re-embed a corpus, how full vs incremental refresh changes spend, and how to amortize reindex cost beside query embeds without inventing live unit prices.
Reindex is a product policy, not a cron default
Embedding reindex cost is mostly how often you pay the full (or partial) corpus token bill again. A cheap $/1M still hurts when you multiply by tens of millions of tokens every night.
Re-embed when content changed enough to stale vectors, when the embedding model or dimensions changed, when chunking strategy changed, or when quality evals show drift you can fix with a refresh. Do not re-embed on a fixed nightly schedule only because the pipeline can. Paste $/1M from the provider pricing page; presets are editable references, not live catalog prices.
Full reindex vs incremental update
Full reindex embeds the entire corpus (with overlap): one large bill per pass, best for model/chunker migrations and rare scheduled rebuilds. Incremental embeds changed or new chunks only: small frequent bills, best for docs that edit daily. Hybrid uses incremental daily plus an occasional full rebuild.
Monthly index API ≈ (tokens embedded in period / 1M) × $/1M. Batch discounts often help full offline passes. Incremental hot updates may stay on standard rates.
How often should I re-embed my corpus?
Mostly static handbook: re-embed on publish (incremental) plus optional quarterly full rebuild. Daily-edited wiki: incremental on change; avoid nightly full. Model upgrade: one planned full reindex with Batch if available. Dimension change: full reindex and budget storage cutover.
There is no universal every-N-days answer. Use retrieval quality and content churn, then price the cadence in the calculator.
Amortizing reindex into monthly views
Estimate corpus tokens, paste live embed $/1M, set full re-embeds per month (for example 0.25 for quarterly, 1 for monthly, 30 for nightly full), add incremental token volume if tracked, and add monthly query embed cost.
For high QPS assistants, generation usually dwarfs index. For huge corpora with tiny traffic and aggressive full reindexes, index can win the share chart. Compare in the RAG cost calculator.
Common mistakes
Avoid these reindex cost mistakes.
- Nightly full reembed on a multi-million-token corpus “to be safe”
- Changing embed model without a reindex budget line
- Ignoring overlap when estimating incremental clusters
- Counting only unique docs while the API sees chunked duplicates
- Forgetting query embeds while obsessing over reindex
- Treating Batch as permission to reindex constantly
Frequently asked questions
- How often should I re-embed my corpus?
- When content, model, dimensions, or chunking change enough to hurt retrieval. Prefer incremental updates; full rebuilds for migrations or rare scheduled refresh.
- Is re-embedding the whole store every night insane?
- For large corpora, usually yes. You pay full corpus tokens × $/1M every night. Use incremental updates instead.
- How is embedding reindex cost calculated?
- Tokens submitted in the refresh / 1M × embed $/1M (Batch rate if applicable). Amortize by refreshes per month in /tools/embedding-cost.
- Does the Batch API reduce reindex cost?
- It can reduce $/1M for eligible offline full passes. It does not fix a bad nightly full schedule.
- How much does it cost to embed a knowledge base once?
- Corpus embed tokens / 1M × live $/1M from the provider pricing page.
- Does overlap increase reindex cost?
- Yes, full and incremental passes embed overlap tokens again. Measure API tokens.
- Is embedding cost or generation cost higher in RAG?
- Often generation at steady query volume. Aggressive reindex can flip the story for low-traffic, huge corpora. Compare in /tools/rag-cost.
- Do I need a full reindex to change metadata only?
- Usually no if vectors are unchanged. Update metadata in the vector DB without re-embedding when the text is identical.
- What if I only change the embedding model?
- Budget a full reindex and storage cutover. Partial mixes of models in one collection usually hurt retrieval.
- Where do I model weekly vs monthly refresh?
- Set re-embed cadence in /tools/embedding-cost with verified rates.
Try it in TokenCalculator
Enter corpus tokens and embed $/1M, set re-embeds per month (or compare weekly vs monthly), and read amortized index cost next to query embeds.
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 · Batch embedding cost savings · 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.