Pure usage pricing. Nothing else.
No clusters. No provisioning. No always-on compute charges. Stored data is billed on Pay as You Go, even when idle.
Start for free
Build prototypes, learn the API, and validate ideas — no credit card required.
- 1 project
- Up to 2 team members
- 2 GB storage
- 2 GB writes / month
- 1 PB reads / month
- 500K LIU inference / month
- Available in 34 AWS regions worldwide
- Community support (Slack)
Scale to billions.
- Pay-as-you-go based on usage
- Available in 34 AWS regions worldwide
- Data versioning: branches, tags, aliases
- Hybrid search + semantic + lexical
- Point-in-time search
Also metered: $0.05 / GB returned (query responses) · $0.10 / GB-mo retention storage (deleted data kept in the retention window). Free for a promotional period, billing starts with advance notice.
Embeddings and reranking, on demand.
$1.05 per 1M LIU, using the per-model rates below, before plan allowances.
Embedding models
Text input rates are shown below. Cohere Embed v5 models also support image inputs.
| Provider | Model | Text input LIU / tokenLIU / tok | ≈ Cost / 1M text tokens$ / 1M tok |
|---|---|---|---|
| OpenAI | OpenAItext-embedding-3-smallText | 0.02 | $0.021 |
| OpenAI | OpenAItext-embedding-3-largeText | 0.13 | $0.1365 |
| OpenAI | OpenAItext-embedding-ada-002Text | 0.1 | $0.105 |
| Cohere | Cohereembed-v5.0-proText + image | 0.12 | $0.126 |
| Cohere | Cohereembed-v5.0-fastText + image | 0.08 | $0.084 |
For Embed v5 Pro and Fast, image inputs are billed separately at 0.4 LIU per image token (approximately $0.42 per 1M image tokens). Image tokens are reported by the provider.
Reranking models
| Provider | Model | Input LIU / tokenLIU / tok | ≈ Cost / 1M input tokens$ / 1M tok |
|---|---|---|---|
| TypeSafe | TypeSafejev-1.13.0 | 0.042 | $0.0441 |
| Cohere | Coherererank-v5.0-pro | 0.05 | $0.0525 |
| Cohere | Coherererank-v5.0-fast | 0.02 | $0.021 |
Reranking output tokens add no LIU. Native query embeddings and query reads are billed separately.
Free tier includes 500K LIU / month. How LIU is calculated →
Pay as You Go estimate; Free plan allowances are not deducted.
Compare Free and Pay as You Go
Every core feature is available on every plan. Pay as You Go unlocks unlimited scale.
| Feature | Free | Pay as You Go |
|---|---|---|
| Hybrid search (BM25 + multi-vector) | ||
| Flexible document model | ||
| Full-text search (BM25) | ||
| Filtered search | ||
| Range search | ||
| Multi-language FTS (EN, KO, JA) | ||
| Vector compression | ||
| Automatic sharding | ||
| Git-like branching | ||
| Partitioning / multi-tenancy | ||
| Max projects | 1 | 5 (or higher on request) |
| Max team members | 2 | Unlimited |
Data returned and retention storage are free for a promotional period; billing starts with advance notice.
FAQ
Storage is billed at $0.33 / GB-month, prorated to the hour. If you halve your data mid-month, you pay half.
When your usage reaches a Free plan limit — 2 GB storage, 2 GB writes/month, 1 PB reads/month, or 500K LIU inference/month — additional usage is restricted. You can choose to upgrade to Pay as You Go, with per-unit pricing ($0.33 / GB-month storage, $1.00 / GB writes, $5.00 / PB reads, $1.05 / 1M LIU inference) and no minimum charge.
LIU (LambdaDB Inference Unit) is the unified billing unit for native embeddings and reranking. Token cost varies by model, so we normalize usage at a single rate of $1.05 per 1M LIU. 500K LIU on Free ≈ 25M input tokens of text-embedding-3-small, ≈ 3.8M of text-embedding-3-large, or ≈ 5M of text-embedding-ada-002.
Partitioning prevents your queries from scanning the entire index. LambdaDB reads only the partitions you choose, so you pay for fewer bytes scanned and get lower latency. Same index, same accuracy — just faster and cheaper.
LambdaDB uses a serverless, disaggregated architecture, so you don't manage clusters yourself. When traffic spikes, we scale up transparently and you only pay for resources actually used — no pre-provisioning for the peak of the peak, no permanently oversized clusters.
LambdaDB scales to billions of vectors in a single logical index at production-scale latency. Sharding and redistribution happen automatically — no manual re-sharding when your dataset grows.
Start small. Experiment safely.
Scale what works.
Hybrid search with no clusters to size. Branch, evaluate, and promote with aliases.