Pinecone Pricing 2026: Every Plan, Real Costs and Hidden Fees
Pinecone pricing in 2026: free Starter, Builder at $20 flat, Standard with a $50 monthly minimum, read and write unit rates, and what a 10M-vector RAG app costs.

Pinecone is free on Starter, $20 a month flat on Builder, and usage-billed on Standard with a $50 monthly minimum, or $500 on Enterprise. There is no per-seat fee: Starter allows 2 users, Builder 5, and Standard unlimited. On Standard, storage costs $0.33 per GB a month, reads $16 to $18 per million read units and writes $4 to $4.50 per million write units, depending on cloud and region, read on Pinecone's pricing page on November 4, 2026.
The rate card is short, but the bill is not, because a read unit is not a query. Pinecone charges a query by the size of the namespace it searches, so the same app can cost $72 or $381 a month depending on how its data is split. This page prices all four plans, then works through a 10-million-vector RAG app to show where the money goes.
All figures are USD, from Pinecone's pricing page and its cost documentation, both read November 4, 2026.
Toolradar data: of the 20 vector databases tracked on Toolradar, the software directory we run, 16 have a free plan, and Pinecone is listed as an alternative to 23 other tools, third of the 20, as of November 2026. Its record is Pinecone on Toolradar.
| Plan | Price | Storage | Write units | Read units | Users | Who it fits |
|---|---|---|---|---|---|---|
| Starter | $0 | Up to 2 GB | Up to 2M a month | Up to 1M a month | Up to 2 | Prototypes on AWS us-east-1 |
| Builder | $20 a month, flat | Up to 10 GB | Up to 5M a month | Up to 2M a month | Up to 5 | Solo developers who want a fixed bill |
| Standard | $50 monthly minimum, then usage | $0.33 per GB a month | $4 to $4.50 per million | $16 to $18 per million | Unlimited | Production apps of any size |
| Enterprise | $500 monthly minimum, then usage | $0.33 per GB a month | $6 to $6.75 per million | $24 to $27 per million | Unlimited | Teams needing a 99.95% SLA, BYOC, audit logs |
How Pinecone bills: four meters and a minimum
Every serverless index is metered on read units, write units, storage and egress. Storage is the total size of your records: ID, metadata and 4 bytes per vector dimension. Egress is free up to 100 GB a month on Standard and Enterprise, then $0.10 per GB. The other two meters are where estimates go wrong.
A query costs 1 read unit for every 1 GB of the namespace it searches, with a minimum of 0.25. The top_k value and whether you return metadata do not change the price; only the namespace size does. A fetch costs 1 read unit per 10 records, and a list request 1 read unit. A write costs 1 write unit per KB of the request, with a minimum of 5, and an upsert that overwrites a record also pays for the KB of the record it replaces, so re-embedding a corpus costs roughly twice what loading it did.
The minimum is a floor, not a fee on top. On Standard, a month with usage below the minimum is billed the full minimum, with a line item covering the gap, and a month with $100 of usage is billed $100. Builder works the other way: the flat fee is the whole bill, and usage past Builder's limits is blocked rather than charged. That makes Builder the safer plan for a side project with unpredictable traffic, and the wrong plan for anything that must keep serving when it hits a ceiling.
What a 10-million-vector RAG app costs, computed
Pinecone's own docs size an index of 10 million records with 1,536-dimension embeddings and 1,000 bytes of metadata at 71.5 GB. That is a realistic RAG corpus: a few hundred thousand documents chunked into passages, embedded with a 1,536-dimension model. It is far past Builder's 10 GB, so it needs Standard.
Fixed parts of the monthly bill, our arithmetic on Standard's printed rates:
- Storage: 71.5 GB at $0.33 is $23.60 a month.
- Writes: adding 500,000 new chunks a month in batches of 1,000 uses about 3.57 million write units (Pinecone's docs put a 1,000-record batch of this shape at 7,140 units), or $14.28 to $16.07.
The read bill depends on query volume and on how the data is split. In one namespace, every query searches the whole index and costs one read unit per GB of it. Split the same data into 10 namespaces of about 7.15 GB, one per customer or per product line, and a query that targets one namespace costs 7.15 read units.
| Queries a month | One 71.5 GB namespace: reads | Ten 7.15 GB namespaces: reads | Monthly total, one namespace | Monthly total, ten namespaces |
|---|---|---|---|---|
| 30,000 (1,000 a day) | $34.32 to $38.61 | $3.43 to $3.86 | About $72 to $78 | $50, the minimum (usage about $41 to $44) |
| 300,000 (10,000 a day) | $343.20 to $386.10 | $34.32 to $38.61 | About $381 to $426 | About $72 to $78 |
| 3,000,000 (100,000 a day) | $3,432 to $3,861 | $343.20 to $386.10 | About $3,470 to $3,901 | About $381 to $426 |
Totals add storage and 500,000 monthly writes, on Standard, before egress, embedding and tax. The low end is the cheapest region on Pinecone's range, the high end the dearest.
Namespace design cuts the read bill by ten at every volume. That only works when each query belongs to one slice of the data: a multi-tenant SaaS where every customer searches their own documents is the perfect case, and a single global search box is not. If your app genuinely needs to search all 10 million chunks at once, the single-namespace column is your bill, and above a few hundred thousand queries a month, reads are nearly the whole of it.
At high, steady volume Pinecone sells Dedicated Read Nodes on Standard, provisioned capacity with no read-unit limits for queries on that index. Check the current price on Pinecone's site before you compare them with the read-unit totals above.
Devshot is a free daily email for developers that flags pricing changes on infrastructure like Pinecone the week they happen. Join at devshot.co.
The one-time costs: loading and embedding
Loading the corpus by upsert costs about $286 to $321 in write units. At 7,140 write units per 1,000-record batch, 10 million records use 71.4 million write units. The same index loaded through import from object storage, a Standard and Enterprise feature at $0.25 per GB, costs about $17.88, our arithmetic. For an initial load or a full rebuild, staging the vectors in S3 or another object store and importing them is the cheapest move on the price list; our object storage pricing comparison prices the staging bucket.
Embedding is a separate meter if Pinecone does it for you. Pinecone Inference charges $0.16 per million tokens for llama-text-embed-v2 and $0.08 for multilingual-e5-large on Standard. If your 10 million chunks average 400 tokens, embedding them with llama-text-embed-v2 is 4 billion tokens, or $640, our arithmetic on that assumption. Bringing embeddings from another provider moves that line to that provider's bill; our OpenAI API pricing, Gemini API pricing and Claude API pricing explainers cover the model side of a RAG bill, and free LLM APIs lists the no-cost options.
The fees people miss
- The Standard minimum on a quiet app. A small production index that uses $10 of resources still pays the full Standard minimum, $600 a year. Builder's flat fee costs $360 less over 12 months, if the data fits its 10 GB and traffic stays under its limits.
- Overwrites. An upsert that replaces a record pays for both the new and the old KB, so a monthly re-embedding of the whole 10-million corpus costs roughly double the initial upsert load.
- Backups. $0.10 per GB a month on Standard, so one backup of this index adds $7.15 a month; restoring costs $0.15 per GB, and a cross-region restore adds transfer pricing.
- Reranking. $2 per 1,000 requests on Standard for hosted rerankers such as bge-reranker-v2-m3 and pinecone-rerank-v0. A RAG app that reranks every query at 300,000 queries a month adds $600, our arithmetic.
- HIPAA. A $190 monthly add-on on Standard; included on Enterprise.
- Enterprise rates. Read and write units cost 50% more per million on Enterprise than on Standard, on top of the $500 minimum, so the upgrade is for the SLA and controls, not the price.
- Support. Response SLAs are a Developer or Pro support add-on on Builder and Standard; Enterprise includes Pro support.
Pinecone also sells prepaid credits from $8,000 to $25,000, which add extra usage capacity at no extra cost, with benefits that vary by plan and amount.
What changed since 2025
An archived copy of Pinecone's pricing page from January 15, 2025 showed three plans, Starter, Standard and Enterprise, with Standard sold as pay as you go and no monthly minimum on its card. The same copy listed import from object storage in public preview at $1 per GB. The read and write unit rates started at the same $16 and $4 per million they start at now.
So the November 2026 ladder adds two things: the flat-fee Builder plan, which fills the gap between free and production, and the $50 Standard floor, which makes very small production indexes more expensive than they were. Import, on the other hand, is a quarter of its preview price.
Pinecone against the alternatives
Pinecone's model rewards apps with partitioned data and modest query volume, and punishes a single huge namespace under heavy traffic. Open-source engines that you host yourself charge for the server, not the query, so a busy global search can cost less on a fixed instance; our free vector databases list covers the self-hosted and free-tier options, and if you already run Postgres, the pgvector route in our Supabase pricing and Neon pricing explainers keeps vectors in the database you pay for anyway. MongoDB Atlas pricing covers the same idea for document stores.
For a ranked comparison of managed options, Aipresso's best vector databases guide sets Pinecone next to Weaviate, Qdrant, Chroma and Cloudflare Vectorize.
Which Pinecone plan should you pay for?
Stay on Starter for a prototype under 2 GB, about 280,000 vectors of this shape, if AWS us-east-1 and 2 users are enough. Our cloud free tiers comparison shows what else you can run at no cost alongside it.
Pick Builder for a solo project or small team whose index fits 10 GB, roughly 1.4 million of these 1,536-dimension vectors by our arithmetic, and who prefer a hard stop to a surprise bill.
Pick Standard for production, any index above 10 GB, or any app that must keep answering past Builder's limits. Use the 3-week trial and its $300 of credits to measure real read units per query before you commit, and design namespaces around how users search.
Pick Enterprise only for the 99.95% uptime SLA, bring-your-own-cloud, private endpoints, customer-managed keys or audit logs; the higher unit rates make it the expensive way to buy capacity.
How we checked these prices
We read Pinecone's pricing page, including the plan comparison table, and its cost documentation on November 4, 2026. Index size, write units per batch and read units per query follow Pinecone's own formulas and examples; every total is our arithmetic on the printed rates for a 30-day month, shown as a range because rates vary by cloud and region. The 400-token chunk size is an assumption for illustration, not a benchmark. The historical prices come from the archived January 2025 copy of the same page. Prices change, so check the pricing page before you commit.
Cite this: Devshot, "Pinecone Pricing 2026: Every Plan, Real Costs and Hidden Fees", November 2026.
FAQ
How much does Pinecone cost in 2026?
Pinecone has four plans as of November 4, 2026: Starter is free, Builder costs $20 a month flat, Standard bills usage with a $50 monthly minimum, and Enterprise bills usage with a $500 minimum. On Standard, storage is $0.33 per GB a month, read units $16 to $18 per million and write units $4 to $4.50 per million, depending on cloud and region.
Does Pinecone charge per seat?
No. Pinecone charges for usage, not users. Starter allows up to 2 users and Builder up to 5, while Standard and Enterprise allow unlimited users at no extra cost. A team of 20 on Standard pays the same as a team of 2 with the same index and traffic.
How much does a 10-million-vector index cost on Pinecone?
With 1,536-dimension vectors and 1 KB of metadata, the index is 71.5 GB, or $23.60 a month of storage on Standard. Reads dominate: 300,000 queries a month against one namespace cost $343.20 to $386.10, but about a tenth of that if the data is split into ten namespaces that each query targets, by our arithmetic.
Is Pinecone free?
Yes, on Starter: up to 2 GB of storage, 2 million write units and 1 million read units a month, 5 indexes and 2 users, on AWS us-east-1 only. Standard also comes with a 3-week trial that includes $300 of credits.
What is the difference between Pinecone Builder and Standard?
Builder is a flat $20 a month with caps of 10 GB storage, 5 million write units and 2 million read units, and usage past the caps is blocked. Standard has no caps and bills usage with a $50 minimum, and adds import from object storage, backups, Dedicated Read Nodes, SSO and the HIPAA add-on.
Why is my Pinecone read bill so high?
Because a query costs 1 read unit per GB of the namespace it searches, regardless of top_k. A query against the 10-million-vector namespace in our example costs 71.5 read units, ten times what it costs against a tenth of the data. Splitting data into smaller namespaces per tenant, when each query only needs one, is the largest saving available.
Devshot sends the plan changes and price moves behind your stack in one short daily email. Subscribe at devshot.co.
Some offers on this page may be paid placements or contain affiliate links.
Devshot: the daily dev & AI brief
Free daily newsletter, read in 5 minutes.
Subscribe free