Turbopuffer Pricing in 2026 (vs Pinecone)

Checked on 2026-10-19 against the sources cited below.

Turbopuffer stores vectors on object storage and bills mainly for the data each query scans, which makes it cheap for both small namespaces (they fall under the minimum) and large ones (scanned data is discounted heavily above 32 GB). This article sets out the rates, what changed in 2026, and how it compares with Pinecone on the same workloads.

1. How Turbopuffer prices: storage, writes and data queried

Turbopuffer publishes its usage rates in the cost calculator on its pricing page^[1]. On the Launch and Scale plans:

• Storage: $0.33 per GB-month of logical size; a vector dimension counts 4 bytes as float32.
• Writes: $2 per GB written, with a 10 KB minimum per write and a discount of up to 50% for large batches.
• Queries: billed on data queried, the size of the namespace each query searches, at least 1.28 GB per query, at $1 per petabyte, with an 80% marginal discount between 32 GB and 128 GB and 96% beyond 128 GB^[2].
• Data returned: $0.05 per GB.

The Embeddings DB Cost calculator uses these rates, with float32 vectors, unbatched writes and 10 results of about 1 KB returned per query, then applies the plan minimum^[4].

2. The $16 minimum and what changed in 2026

Turbopuffer's changelog records the changes that matter for a small team^[2]:

• June 2026: the Launch minimum fell from $64 to $16 a month, for new and existing customers. A month with less than $16 of usage bills $16; above that you pay usage.
• February 2026: the base rate for data queried fell from $5 to $1 per petabyte, with the 80% and 96% discounts for large namespaces and a 1.28 GB minimum per query.
• April and August 2026: namespace pinning (billing a namespace held in memory by GB-hour instead of per query) arrived with a 64 GB minimum, raised to 128 GB in August. It applies only to pinned namespaces.

So a small unpinned namespace does not carry a size floor: its cost is its usage or the $16 minimum, whichever is higher.

3. Turbopuffer vs Pinecone on the same workload

Two scenarios through the Embeddings DB Cost engine: a small RAG app (200,000 vectors at 1,536 dimensions, 2,400 queries and 1,500 ingests a day) and a large one (20 million vectors, 100,000 queries and 10,000 ingests a day).

At 200,000 vectors both vendors bill their minimums: Turbopuffer $16, Pinecone $50^[1][3]. At 20 million vectors the difference is the query model. Pinecone charges a read unit per GB of namespace on every query, so 3 million queries a month over a 143 GB index come to about $6,921. Turbopuffer counts about 50 GB of billable data per query after its large-namespace discounts, at $1 per petabyte, which with storage and writes comes to about $196.

4. The verified comparison table

Dimension	Turbopuffer	Pinecone
Architecture	Object-storage-native	Managed serverless[3]
Entry plan minimum	$16 / month (Launch)[1]	$50 / month (Standard)[3]
Higher tier	$256 / month (Scale)[1]	Usage scales above minimum[3]
Query billing	$1 per PB of data queried (namespace size per query, discounted above 32 GB)[2]	$16-$18 per million read units; a query uses 1 per GB of namespace[3]
Storage	$0.33 / GB-month[1]	$0.33 / GB-month[3]
Free tier	None (plan minimum)[1]	Starter: 2 GB, 1M read and 2M write units[3]

5. When Turbopuffer is the cheapest option

1. A small production app that needs hosting: $16 a month is below Pinecone Standard's $50 and the $125 a Supabase Postgres sized to hold 200,000 such vectors in memory costs, with nothing to operate.
2. Large namespaces queried often: the per-petabyte query rate and its large-namespace discounts keep the bill far below per-read-unit pricing as the index grows.
3. Not when it must be free: for a prototype, Pinecone's Starter plan or Qdrant's free cluster cost nothing, and LanceDB on object storage costs cents if you run it in your app.

For the full ranking, the cheapest vector database article prices eight vendors at one scenario, the Chroma vs LanceDB article covers the object-storage OSS path, and the Pinecone vs Qdrant article covers the free-tier alternative.

Frequently asked questions

How much does Turbopuffer cost in 2026?

Turbopuffer's Launch plan has a $16 monthly minimum (cut from $64 in June 2026), Scale $256 a month and Enterprise from $4,096 a month. Above the minimum you pay for usage: $0.33 per GB-month of logical storage, $2 per GB written, $1 per petabyte of data queried (each query counts the namespace size, at least 1.28 GB, with discounts above 32 GB and 128 GB) and $0.05 per GB returned, per the rates in its pricing calculator. A small RAG app's usage stays under $16, so it pays the minimum.

Is Turbopuffer cheaper than Pinecone?

On the workloads this site prices, yes. At 200,000 vectors with 2,400 queries a day, Turbopuffer bills its $16 minimum and Pinecone Standard its $50 minimum. At 20 million vectors with 100,000 queries a day, the Embeddings DB Cost engine puts Turbopuffer at about $196 a month and Pinecone at about $6,921, because Pinecone charges each query a read unit per GB of namespace while Turbopuffer charges about $1 per petabyte queried, discounted for large namespaces.

What is the cheapest serverless vector database in 2026?

Among hosted services with a published price, Turbopuffer's $16 Launch minimum is now the lowest paid entry this site tracks, and a free tier such as Qdrant's free cluster or Pinecone's Starter costs nothing for a prototype. An open-source engine on your own object storage, such as LanceDB on Cloudflare R2, has the lowest bill of all (cents a month for a small app) but you run it inside your application.