turbopuffer review
by turbopuffer Inc.Vector databases
Serverless vector and full-text search on object storage, billed per GB stored, written and queried, from a $16 monthly minimum, with no free tier.
- Paid only
- Self-serve
- Usage-based
- Public prices
What you're paying for
- Worth the price (main value)
- Saves engineering time
Vector and full-text search billed by the bytes you store, write and query, with every database feature on the $16 plan.
turbopuffer keeps namespaces on object storage and bills logical bytes rather than nodes or replicas, so a large index that is queried occasionally costs little to keep. Our scenario comes to $30 a month on launch, the lowest of the five products in this category. Hybrid BM25 and vector search, filtering, unlimited namespaces, native embedding models and EU regions are on every plan; scale adds SSO, audit logs and a HIPAA BAA from a $256 minimum.
- Billed by bytes, not capacityStorage is $0.33 per logical GB-month, writes $2 per GB and queries $1 per PB searched, with marginal discounts of 80% above 32 GB and 96% above 128 GB per query.Evidence: Pricing page calculator and FAQ (2026-10-16)
- Every feature on the first planThe pricing page says each plan "Includes all database features": vector and BM25 search with server-side rank fusion, filtering and unlimited namespaces.Evidence: Pricing page and hybrid search docs (2026-10-16)
- Native embedding modelsEmbedding models such as openai/text-embedding-3-small ($0.02 per million tokens) and voyage-4 ($0.06) run inside turbopuffer, billed per token.Evidence: Embedding docs (2026-10-16)
- EU and US regions on every planAWS and GCP regions in the US and Europe (Frankfurt, Ireland, Belgium and London) are listed with no plan restriction, and data stays in the region you choose.Evidence: Regions and security docs (2026-10-16)
- Best for
- Teams with large or many namespaces that query each one occasionally and want to pay for bytes rather than provisioned capacity.
- Good to know
- No free tier or trial, no automated backups, no bulk import from object storage and no open-source engine.
How turbopuffer compares
Against 3 other products we reviewed in Vector databases: what you pay for with each, the five access facts, the plan limits we recorded and each review's bottom line. The note under the table says how each one was checked.
| Metric | turbopuffer | Zilliz Cloud | Qdrant Cloud | Pinecone |
|---|---|---|---|---|
| What you're paying for | Worth the priceVector and full-text search billed by the bytes you store, write and query, with every database feature on the $16 plan. | Saves engineering timeManaged Milvus with a serverless option billed per request, free bulk import from object storage, and the Apache 2.0 engine as a way out. | Worth the priceA managed open-source vector engine on nodes you size, with no per-query charge, an exit to self-hosted Qdrant and a free cluster for about 1M vectors. | Saves engineering timeA serverless vector index with hybrid search, hosted models, backups and bulk import on a plan bought online, with no servers or replicas to size. |
| Overall score | 5.6 | 7.2Best | 7.0 | 6.8 |
| Free option | Paid only | Free plan | Free plan | Free plan |
| Card to start | –Not measured | No card needed | No card needed | –Not measured |
| Getting access | Self-serve | Self-serve | Self-serve | Self-serve |
| Starter priceCheapest paid plan | $16/mo minimum | $4/million vCUs | $0.03566/node-hour | $20/mo |
| Pricing model | Usage-based | Usage-based | Usage-based | Usage-based |
| Pricing page | Public prices | Public prices | Public prices | Public prices |
| Billing unit | GB stored, GB written, PB queried and GB returned | Serverless: vCUs and GB-month; Dedicated: CU-hours and GB-month | Provisioned nodes per hour (RAM, vCPU and disk), plus backup storage | Read units, write units and GB-month of storage (serverless) |
| Free option | None; refund if cancelled within 30 days (sign-up page) | Free 5 GB cluster, no card; $100 trial credits for 30 days | Free 1 GB cluster (about 1M 768-dimension vectors), no card | Starter free tier (2 GB, AWS us-east-1); 21-day Standard trial with $300 in credits |
| Reference scenario (1M vectors, 1M queries) | $30/mo on launch, computed | $113.28/mo on Serverless (GCP us-west1), computed | $102.51/mo on Standard (3 × mx1 nodes), official calculator | $176.60/mo on Standard, computed |
| Plan minimum | launch $16; scale $256; enterprise from $4,096 a month | None | None on Standard; Premium minimum not published | Builder $20 flat; Standard $50; Enterprise $500 a month |
| Hybrid search | Every plan (BM25 plus vector, rank fusion) | Every plan (BM25 full-text plus vector) | Every plan (dense plus sparse or BM25) | Every plan (dense, sparse and full-text indexes) |
| Backups | Manual namespace copies and branches; no automated backups | Dedicated clusters; point-in-time recovery on Business Critical | Standard and up, scheduled; collection snapshots on every plan | Standard and up, scheduled; $0.10 per GB-month |
| SOC 2 / ISO 27001 | SOC 2 Type II; ISO 27001 not stated | SOC 2 Type II (under NDA) and ISO 27001:2022 | SOC 2 Type II; ISO 27001 not stated | SOC 2 Type II and ISO 27001:2022 |
| HIPAA BAA | scale and enterprise | Business Critical, through sales | From Standard; BAA scope through Solutions Engineering | Standard add-on $190/mo (6-month minimum); included on Enterprise |
| EU regions | AWS Frankfurt, Ireland and London; GCP Frankfurt and Belgium | Serverless in AWS Frankfurt; Dedicated in AWS, GCP and Azure EU regions | AWS Frankfurt and Ireland, GCP Frankfurt, Azure Germany West Central | AWS Ireland and Frankfurt, GCP Netherlands, from Builder |
| SSO | scale and enterprise, enabled on request | SAML or OIDC on Enterprise and up | Premium (console sign-in) | SAML on Standard and Enterprise |
| BYOC | enterprise | Through sales, annual commitment | Hybrid Cloud and Private Cloud, through sales | Enterprise only, priced by sales |
| Data export | Paged query API (JSON) | SDK iterators; backup file export in private preview | Downloadable snapshots; scroll API | List and fetch API; no bulk export |
| Self-host path | None | Milvus (Apache 2.0), same SDKs | Qdrant open source (Apache 2.0), restore from snapshot | None (Pinecone Local reached end of life on Oct 7, 2026) |
| Bottom line | From its public pages, turbopuffer is the lowest-cost option for our scenario, with every database feature from a $16 minimum. It has no free tier, no automated backups and no open-source engine, and some pricing details differ between pages. | From its public pages, Zilliz Cloud is managed Milvus with per-request serverless billing, free bulk import, SOC 2 Type II and ISO 27001, and an open-source exit. Serverless runs in two regions only, and several prices differ between its pages. | From its public pages, Qdrant Cloud is managed open-source Qdrant billed by node-hour: no per-query charge, a large free cluster and a documented way back to self-hosting. Prices live in the calculator, and SSO and private networking need Premium. | From its public pages, Pinecone puts the most on a self-serve plan here: hybrid search, backups, bulk import, SAML SSO and a HIPAA add-on all on Standard. Query cost grows with namespace size, and there is no open-source engine to fall back on. |
Evidence: turbopuffer: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Zilliz Cloud: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Qdrant Cloud: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run; Pinecone: facts checked 2026-10-16 from its public sign-up and pricing pages, no completed hands-on run. Prices in USD as listed, excluding tax. "–" means not recorded.
Score
- Pricing clarity
- 4.0 / 10, Fair
- unit prices published; can be bought online; month to month; minimum shown on the pricing page or none; full-text attributes are billed "at the same rate" in the FAQ, at up to 200% of their size in the write docs; the sign-up page offers "Cancel within 30 days for a refund", while the terms call fees non-refundable; SLAs are listed for scale on the sign-up page, for enterprise only on the pricing page; the enterprise "35% usage premium" is not applied to storage in the calculator; reranking and some embedding models ("pricing is TBD") have no published price; Azure eastus2 is listed as a region in the docs, "private preview" on the roadmap
- Free tier and trial
- 0.0 / 10, Fair
- no free tier or trial; the sign-up page offers a refund if you cancel within 30 days
- Price for the reference scenario
- 9.6 / 10, Excellent
- $30/mo on launch; storage 7.14 GB × $0.33 = $2.36; writes 1M × 10 kB (the per-write minimum) = 10 GB × $2 = $20; queries 1M × 7.14 GB queried = 7.14 PB × $1 = $7.14; returned 1M × about 10 KB = 10 GB × $0.05 = $0.50; above the $16 minimum; assumes about 10 KB returned per query (10 IDs with their metadata)
- Search and data features
- 8.5 / 10, Excellent
- hybrid search on the scenario plan; metadata filtering on the scenario plan; multi-tenancy on the scenario plan; backups on the scenario plan; integrated embedding or reranking on the scenario plan; bulk import from object storage not listed
- Compliance and deployment
- 5.5 / 10, Fair
- SOC 2 Type II on the scenario plan; no ISO 27001 stated; HIPAA BAA on a higher self-serve plan; EU region on the scenario plan (AWS eu-central-1 and eu-west-1, GCP europe-west3 and europe-west1); BYOC enterprise or through sales only; customer-managed keys enterprise or through sales only; private networking enterprise or through sales only
- Access control and exit
- 6.1 / 10, Good
- SSO on a higher self-serve plan; RBAC or scoped keys on the scenario plan; audit log on a higher self-serve plan; full export documented; no open-source engine with the same API
Review
Review from public pages
Written from turbopuffer's public pages (pricing, security or privacy, and docs) as read on Oct 16, 2026. We did not sign up, and no hands-on run is published yet.
What the public pages show
We read turbopuffer's pricing page with its cost calculator (rendered in a browser, since prices fill in by script), the sign-up page, the docs on writes, queries, hybrid search, limits, regions, backups, branching, pinning, export, embedding, security, audit logs, private networking, encryption and enterprise deployments, the pricing log and roadmap, the terms of service and the DPA on Oct 16, 2026; we did not sign up or call the API. The trust center sits behind a bot check and did not load for us.
Pricing is in bytes. Each write is billed at least 10 kB, so a single 7.14 KB upsert counts as 10 kB, and larger batches earn up to 50% off. Each query is billed for the size of the namespace it searches, at least 1.28 GB, so many small namespaces cost less to query than one large one. The unit prices appear only in the calculator and its tooltips.
Some pages disagree. The FAQ bills full-text attributes "at the same rate" while the write docs bill a filterable, full-text attribute at 200%; and the sign-up page offers "Cancel within 30 days for a refund" while the terms call fees non-refundable. The pricing FAQ also says "We don't currently offer migration tools."
Strengths
- Lowest scenario cost in this category
Our scenario comes to $30 a month on launch: $2.36 storage, $20 writes, $7.14 queries and $0.50 returned data.
vs Qdrant Cloud: Qdrant Cloud comes to $102.51 and Pinecone to $176.60 for the same scenario.
- All features on every plan
Hybrid search with rank fusion, filtering, unlimited namespaces and native embedding are on launch, from a $16 minimum.
vs Pinecone: Pinecone puts backups and bulk import on Standard, above Builder.
- EU regions without a plan restriction
Frankfurt, Ireland and London on AWS, and Frankfurt and Belgium on GCP, are listed for every plan.
Compared: Weaviate Cloud ties region selection for data residency to Dedicated tiers.
- SSO and HIPAA on a self-serve plan
scale, from a $256 minimum, adds SSO, audit logs (streaming for $128 a month), a HIPAA BAA and IP allowlisting, switched on by contacting turbopuffer.
Things to know
- No free tier or trial
The FAQ describes a "commercial-only model": the cheapest option is launch at a $16 monthly minimum.
Workaround: The sign-up page offers a refund if you cancel within 30 days.
- Backups are manual copies
The docs say "We don't currently offer automated backups"; you copy a namespace with copy_from_namespace or branch it, billed as writes and storage.
- Exit is a paged query
Export pages through documents with the query API; there is no file or Parquet export, no migration tooling and no open-source engine.
- Enterprise options through sales
BYOC, single-tenancy, private networking and customer-managed keys are on enterprise, from $4,096 a month with a 35% usage premium.
Pricing reality
turbopuffer bills logical bytes on every plan: $0.33 per GB-month stored, $2 per GB written (at least 10 kB per write; up to 50% off for batches up to about 3 MB), $1 per PB queried (at least 1.28 GB per query) and $0.05 per GB returned, above a $16 monthly minimum on launch or $256 on scale. For our scenario: storage 7.14 GB × $0.33 = $2.36; 1M single upserts billed at the 10 kB minimum, 10 GB × $2 = $20; 1M queries × 7.14 GB = 7.14 PB × $1 = $7.14; about 10 GB returned, $0.50: $30 a month (computed). Writes dominate here, and batching cuts them. Costs rise with pinned namespaces (about $9.67 per GB a month in the docs’ calculator, 128 GB minimum), audit log streams ($128), the enterprise premium and native embedding by token.
Exit path
Documents leave through the query API, paging by advancing a filter on the id attribute, with vectors and attributes returned as JSON; native embedding vectors can be read and exported too. There is no file export, no migration tool and no open-source engine, and a deleted namespace cannot be recovered. On termination the terms say customer data is deleted; the DPA commits to deletion within 30 days of a request.
Compared with
- Pick turbopuffer when
- You want the lowest bill for our scenario ($30 against $176.60) and every feature from a $16 minimum.
- Pick Pinecone when
- You want a free tier, scheduled backups and bulk import from object storage.
- Pick turbopuffer when
- You want to pay for bytes rather than size nodes, with unlimited namespaces.
- Pick Qdrant Cloud when
- You want a free cluster, an Apache 2.0 engine to self-host and downloadable snapshots.
- Pick turbopuffer when
- You want serverless pricing in AWS US regions such as us-east-1.
- Pick Zilliz Cloud when
- You want a free tier, free import from object storage and an open-source engine.
Choose it if
- You keep large or many namespaces and query each one occasionally.
- You want hybrid search and native embedding without a plan upgrade.
- You batch writes and want to pay by bytes, not provisioned capacity.
Look elsewhere if
- You want a free tier to start, or automated backups.
- You need an open-source engine or a file export as a fallback.
Bottom line
From its public pages, turbopuffer is the lowest-cost option for our scenario, with every database feature from a $16 minimum. It has no free tier, no automated backups and no open-source engine, and some pricing details differ between pages.
Plan limits at a glance
| Limit | As recorded |
|---|---|
| Billing unit | GB stored, GB written, PB queried and GB returned |
| Free option | None; refund if cancelled within 30 days (sign-up page) |
| Reference scenario (1M vectors, 1M queries) | $30/mo on launch, computed |
| Plan minimum | launch $16; scale $256; enterprise from $4,096 a month |
| Hybrid search | Every plan (BM25 plus vector, rank fusion) |
| Backups | Manual namespace copies and branches; no automated backups |
| SOC 2 / ISO 27001 | SOC 2 Type II; ISO 27001 not stated |
| HIPAA BAA | scale and enterprise |
| EU regions | AWS Frankfurt, Ireland and London; GCP Frankfurt and Belgium |
| SSO | scale and enterprise, enabled on request |
| BYOC | enterprise |
| Data export | Paged query API (JSON) |
| Self-host path | None |
Read from the vendor's public pages on Oct 16, 2026. No hands-on run is published yet. How we check
We have not published a sign-up run for turbopuffer yet. The access facts above were read from its sign-up and pricing pages on Oct 16, 2026.
We have not published API calls for turbopuffer yet.
turbopuffer lists 3 plans. The cheapest paid plan, launch, costs $16/mo minimum.
- turbopufferLowest pricelaunch$16/mo minimumUsage above the minimum: $0.33/GB stored, $2/GB written, $1/PB queried, $0.05/GB returned
- All database features: vector, BM25 full-text and hybrid search
- Unlimited namespaces
- SOC 2 report and GDPR-ready DPA
- Community Slack and email support
- turbopufferscale$256/mo minimumSame unit prices as launch
- SSO and audit logs (log streams +$128/mo)
- HIPAA-ready BAA
- IP allowlisting
- Private Slack channel, 8–5 support hours
- turbopufferenterpriseContact usFrom $4,096 a month, with a 35% usage premium
- Single-tenancy and BYOC
- Private networking
- Customer-managed keys per namespace
- 24/7 support, support and uptime SLAs
Prices as listed, USD, excluding tax; observed 2026-10-16.
Source: turbopuffer.com pricing page. Prices change; the observed date above is when our agent read them.