vai estimate

Estimate and compare embedding costs across different models and strategies (symmetric vs. asymmetric retrieval).

Synopsis

vai estimate [options]

Description

vai estimate calculates the total cost of embedding a document corpus and serving queries over time, comparing symmetric strategies (same model for docs and queries) against asymmetric strategies (different models for docs vs. queries).

It uses the Voyage 4 model catalog's pricing to project costs over a configurable time period, helping you choose the most cost-effective model combination.

Options

Flag	Description	Default
--docs <n>	Number of documents (supports K/M/B shorthand)	100K
--queries <n>	Queries per month (supports K/M/B shorthand)	1M
--doc-tokens <n>	Average tokens per document	500
--query-tokens <n>	Average tokens per query	30
--doc-model <model>	Model for document embedding (asymmetric)	voyage-4-large
--query-model <model>	Model for query embedding (asymmetric)	voyage-4-lite
--months <n>	Months to project	12
--json	Machine-readable JSON output	—
-q, --quiet	Suppress non-essential output	—

Examples

Default estimate (100K docs, 1M queries/month)

vai estimate

Large-scale estimate

vai estimate --docs 10M --queries 50M --months 24

Compare specific model combinations

vai estimate --doc-model voyage-4 --query-model voyage-4-lite --docs 500K --queries 5M

JSON output for dashboards

vai estimate --docs 1M --queries 10M --json

Output

Shows a strategy comparison table sorted by total cost:

• Symmetric strategies: Same model for both documents and queries (one entry per Voyage 4 model)
• Asymmetric (recommended): Your chosen doc-model + query-model combination
• Asymmetric with local queries: Doc-model via API + voyage-4-nano for free local queries

Each strategy shows:

• One-time document embedding cost
• Monthly query cost
• Projected total over the specified period
• Savings percentage compared to the most expensive option

Tips

• Numbers support shorthand: 100K, 1M, 10B — no need to type out zeros.
• Asymmetric retrieval works because all Voyage 4 models share the same embedding space. You can embed documents with voyage-4-large for quality and queries with voyage-4-lite for cost savings.
• Use vai explain shared-space to learn more about how asymmetric retrieval works.
• The voyage-4-nano option shows $0 query cost because it runs locally as an open-weight model.

Related Commands

• vai models — View model pricing and specs
• vai embed — Embed with --estimate for per-call cost
• vai pipeline — Estimate costs with --estimate or --dry-run
• vai explain — Learn about shared embedding spaces