> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lambdadb.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Understanding costs

> Learn how LambdaDB usage pricing works across reads, writes, storage, and inference, including request minimums and promotional pricing for data returned and retention storage.

LambdaDB offers a **Free** plan and a **Pay as You Go** plan.

There is **no minimum cost** or base fee. If you have no usage, your cost is **\$0**.

On the **Pay as You Go** plan, reads, writes, storage, and inference are billed from measured usage at the rates below. Data returned and retention storage are also metered, with fees waived during the promotion described below.

For the latest pricing details, see [Pricing](https://lambdadb.ai/pricing).

Use [`lambdadb usage`](https://docs.lambdadb.ai/guides/get-started/use-with-cli#view-usage) to inspect current Free quotas or recorded Pay as You Go usage and cost estimates for a project or organization.

## Plans

LambdaDB offers two plans:

* **Free**: includes up to **1 PB read usage**, **2 GB write usage**, **2 GB storage usage**, and **500K LIU inference usage** per month at **no cost**. After a Free plan limit is reached, additional usage is **restricted**.
* **Pay as You Go**: usage is billed at the rates described below, with no minimum charge or per-collection fixed fee. Inactive collections still incur storage usage while their data is retained.

## Read usage

Read usage is billed at **\$5 per read PB** (petabyte) of measured usage.

The following requests contribute to read usage based on **collection size**:

* **Collection query** ([document query](https://docs.lambdadb.ai/reference/api/endpoint/document-query.md))
* **Document fetch** ([document fetch](https://docs.lambdadb.ai/reference/api/endpoint/document-fetch.md))
* **Document list** ([document list](https://docs.lambdadb.ai/reference/api/endpoint/document-list.md))

For these operations, read usage reflects collection size. On a partitioned collection, `partitionFilter` narrows the request to selected physical partitions, and read usage reflects their combined size. Unselected partitions are not searched. This requires explicit partition selection; an ordinary document predicate does not by itself narrow the billed partition scope. See [Partition selection](https://docs.lambdadb.ai/guides/search/search-overview.md#select-partitions-to-search).

**Per-request minimum:** each request is counted as at least **1 GB** of read usage.

**Strong consistency:** for **query** and **fetch** requests that select a branch directly or through an alias that currently targets a branch, if you set `consistentRead` to `true`, an **additional 1 GB** is added to read usage for that request (on top of the per-request minimum and the collection-size basis). Tag reads and aliases that target a tag do not support strong consistency. See also [Search overview](https://docs.lambdadb.ai/guides/search/search-overview.md) and [Fetch data](https://docs.lambdadb.ai/guides/documents/fetch-data.md).

## Write usage

Write usage is billed at **\$1 per write GB** of measured usage.

**Upsert**, **update**, and **bulk-upsert** ([document upsert](https://docs.lambdadb.ai/reference/api/endpoint/document-upsert.md), [document update](https://docs.lambdadb.ai/reference/api/endpoint/document-update.md), [document bulk upsert](https://docs.lambdadb.ai/reference/api/endpoint/document-bulk-upsert.md)) are measured from the **size in bytes of the document payload** you send in the request body.

**Update** ([document update](https://docs.lambdadb.ai/reference/api/endpoint/document-update.md)) only: each request always runs an internal **document fetch** with strong consistency (equivalent to **fetch** with `consistentRead` set to `true`). That work is billed as **read usage** under [Read usage](#read-usage)—including collection-size basis, the per-request minimum, and the **Strong consistency** rule.

**Per-request minimum** for upsert, update, and bulk-upsert: **50 KB**.

**Delete** ([document delete](https://docs.lambdadb.ai/reference/api/endpoint/document-delete.md)): **100 KB** of write usage is added **per delete request** (fixed amount).

## Storage usage

Storage is billed on **physical storage** at **\$0.33 per GB-month**.

If you use **branching**, logical storage and physical storage can differ; billing follows **physical** storage.

Usage is computed from **hourly averages** of stored data. Those averages are summed into **GB-hours** over the billing period (not a single end-of-month snapshot).

## Data returned and retention storage

The [public pricing page](https://lambdadb.ai/pricing) also lists these usage meters:

| Meter | Published rate | Promotional status |
| :- | :- | :- |
| Data returned in query responses | \$0.05 per GB returned | Free during the promotional period |
| Retention storage for deleted data kept in the retention window | \$0.10 per GB-month | Free during the promotional period |

Billing for these two meters starts with advance notice. Include them when estimating costs beyond the promotion; do not treat the promotional waiver as a permanent free tier. Retention storage for deleted data is distinct from an inactive collection whose documents remain stored. The regular storage charge still applies to that inactive collection.

## Inference usage

Inference usage applies when LambdaDB calls an inference provider on your behalf, such as native embedding generation during document writes, `knn.queryText` or `knn.queryInputs` searches on native embedding fields, or [native reranking](https://docs.lambdadb.ai/guides/search/native-reranking.md). Supplying vectors directly skips embedding inference. A query that also requests reranking still incurs reranking inference usage.

Inference usage is measured in **LIU** (LambdaDB Inference Unit) and billed at **\$1.05 per 1M LIU**.

LambdaDB converts provider/model token usage into LIU using LambdaDB's internal rate card. For example, different models can have different LIU-per-token conversion rates, and image inputs are metered as image tokens with their own rate. Usage is aggregated over the billing period before it is converted to the integer LIU quantity used for billing.

If the aggregated LIU value is positive but less than 1, it is rounded up to **1 LIU** so non-zero native inference usage is not dropped.

See [Native embeddings](https://docs.lambdadb.ai/guides/collections/native-embeddings.md) and [Native reranking](https://docs.lambdadb.ai/guides/search/native-reranking.md) for feature setup and behavior.

### Native embedding rates

| Provider | Model | Text input LIU per token | Image input LIU per token | Approx. cost per 1M text input tokens |
| :- | :- | -: | -: | -: |
| `openai` | `text-embedding-3-small` | 0.02 | — | \$0.021 |
| `openai` | `text-embedding-3-large` | 0.13 | — | \$0.1365 |
| `openai` | `text-embedding-ada-002` | 0.1 | — | \$0.105 |
| `cohere` | `embed-v5.0-pro` | 0.12 | 0.40 | \$0.126 |
| `cohere` | `embed-v5.0-fast` | 0.08 | 0.40 | \$0.084 |

Embedding models have no output tokens. Cohere image inputs are metered in image tokens reported by the provider; one million image tokens equal 400,000 LIU, or **\$0.42**.

### Native reranking rates

| Provider | Model | Input LIU per token | Output LIU per token | Approx. cost per 1M input tokens |
| :- | :- | -: | -: | -: |
| `typesafe` | `jev-1.13.0` | 0.042 | 0 | \$0.0441 |
| `cohere` | `rerank-v5.0-pro` | 0.05 | 0 | \$0.0525 |
| `cohere` | `rerank-v5.0-fast` | 0.02 | 0 | \$0.021 |

At the \$1.05 per 1M LIU rate, one million Jev input tokens equal 42,000 LIU, or **\$0.0441** before plan allowances and integer-LIU rounding. Output tokens add zero LIU. Pricing is based on measured token usage, not provider request count.

A successful reranking stage with complete valid usage records the sum of input/output tokens across its batches. Failed, fallback, or incomplete-usage stages do not produce a customer reranking token event. Query read usage and any native query-embedding usage remain separate from reranking inference usage.

## Billing cycle

On the **first day of each calendar month**, billable usage for the **previous month** is aggregated and invoiced on the **Pay as You Go** plan. The data-returned and retention-storage fees are waived during their promotional period.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.