This guide is based on Pinecone’s official pricing page and documentation as of October 14, 2026 — we did not run the tool hands-on. Every number below was read from the vendor’s pages; we did not benchmark latency, run live queries, or load test the Starter plan ourselves.
What is Pinecone’s free tier in 2026? (The Starter plan, in one paragraph)
Pinecone’s free tier is the Starter plan, which costs $0/month with no monthly minimum, according to the Pinecone pricing page. It is serverless-only (dense, sparse, and full-text indexes) and runs on AWS in us-east-1. Starter allows up to 5 serverless indexes, 100 namespaces per index, 1 project, and 2 users. The pricing page describes it as meant for “trying out and for small applications,” with support limited to the Discord community. If your app stays small, single-region, and solo, Starter is usable at no cost.
Starter plan limits: the full number sheet
Pinecone publishes Starter’s limits in two places: the pricing page summary table and the database object-limits reference. Both are official, and they do not agree on storage (see below). The headline numbers:
- Indexes: 5 serverless indexes per project, all in us-east-1 on AWS (object-limits)
- Namespaces per index: 100 (object-limits)
- Projects and users: 1 project, 2 users (object-limits)
- Storage: 1 GB on the pricing page, but 2 GB of serverless index storage per org on the object-limits page. The two Pinecone pages give different figures, so plan for the lower one until you confirm it in your console.
- Documents: unlimited, per the pricing table
- Input tokens: 500k/month, with 300k output and 500k context tokens listed in the same table
- Ingestion units: 1,000/month, where the pricing table gives 400 tokens (about 300 words) per unit
- Embedding tokens: 5M tokens/month per model for llama-text-embed-v2, multilingual-e5-large, and pinecone-sparse-english-v0 (pricing page)
- Rerank: 500 requests/month on bge-reranker-v2-m3 (pricing page, rate limits)
- Support: community Discord only
The rate-limits reference adds monthly ceilings that are not on the pricing table: 1,000,000 read units and 2,000,000 write units per org per month. These matter once you serve real traffic.
What actually fits on 1 GB and 500k tokens?
The pricing page gives one workload example: a semantic search over about 30K documents averaging 20 KB, with 1024-dimension embeddings, which it estimates at roughly 15,000 searches per day. Pinecone labels this as illustrative, not a quote or a binding offer. Use it as a rough sense of scale, not as capacity planning.
The tightest figure on Starter’s list is the 500,000 input-token monthly allowance. The pricing page groups input tokens with Inference and Assistant usage and does not define what they meter for a vector database, so do not assume it maps to your app’s traffic. The 1,000 ingestion units are likewise undefined for upserts; check your usage in the console before loading a large dataset.
Read units: how queries are metered (and why top_k doesn’t matter)
Serverless reads are billed in read units, not per API call. The understanding-cost docs give the math:
- Query: 1 read unit per 1 GB of namespace size, with a 0.25 RU minimum per query
- Fetch: 1 read unit per 10 records, with a 1 RU minimum per request
- List: 1 read unit, covering up to 100 IDs
- Full-text search: 1 read unit per 1 GB of namespace size, with a 0.25 RU minimum
The key detail from the same page: top_k and include_values do not change query cost, but namespace size does. Returning 10 results or 100 from the same namespace costs the same. Splitting data across namespaces, as the save-on-costs guide recommends for multitenancy, often lowers query cost. For sustained high-query workloads, dedicated read nodes are not subject to read-unit limits, but they are provisioned and billed on a paid plan.
The us-east-1 trap: what single-region, single-cloud means
Starter runs only on AWS in us-east-1, per the pricing page and the object-limits footnote. There is no Azure or GCP option and no region choice on the free tier. That is workable for a U.S.-facing prototype, but it has consequences the limit sheet does not show:
- Every call (embedding, query, upsert) goes to us-east-1 on Starter indexes, which adds latency for users far from that region.
- You cannot colocate your index with an app hosted on Azure, GCP, or another AWS region on the free tier.
For most indie hackers this stays invisible until the first user far from the U.S. East Coast. At that point the paid upgrade is the fix, because Builder is the first tier that allows other clouds and regions.
Legacy pod indexes: don’t build on them
Pinecone’s pod-based indexes guide marks pod indexes as legacy. Standard and Enterprise customers who signed up on or after August 18, 2025 cannot create them. Starter supports serverless indexes, which is the index type the pricing table lists for the free plan.
When to upgrade: Starter → Builder → Standard decision points
The pricing page and the object-limits reference make the upgrade path straightforward:
- Upgrade to Builder ($20/month flat) when you need a second cloud (Azure or GCP), a region other than us-east-1, more than 5 indexes, more than 2 users, more storage, or more than 500k input tokens. Builder allows 10 indexes per project, 5 projects, 5 users, 2M input tokens a month, and 10M embedding tokens per model. Builder storage is also listed differently by the two pages: 3 GB on the pricing page and 10 GB on the object-limits reference.
- Upgrade to Standard ($50/month minimum, pay-as-you-go above) when you outgrow Builder. Standard allows 20 indexes per project, unlimited users, and unlimited storage at $3/GB/month. Its HIPAA add-on is $190/month with a 6-month minimum, per the understanding-cost docs.
Builder has no usage overages: the upgrade guide says it is a flat plan with higher quotas, so Builder usage is blocked at its limits rather than billed.
Pinecone Starter vs Builder vs Standard for a bootstrapper
| Limit | Starter (free) | Builder ($20/mo) | Standard ($50/mo min) |
|---|---|---|---|
| Price model | $0, no monthly minimum | $20/mo flat | $50/mo minimum, pay-as-you-go above |
| Clouds / regions | AWS, us-east-1 only | AWS, Azure, GCP; all regions | AWS, Azure, GCP; all regions |
| Indexes | 5 | 10 per project | 20 per project |
| Projects / users | 1 / 2 | 5 / 5 | 20 / unlimited |
| Storage | 1 GB (pricing) or 2 GB (object-limits) | 3 GB (pricing) or 10 GB (object-limits) | Unlimited, $3/GB/mo |
| Rerank | 500 requests/mo | 1,000 requests/mo | Not stated in sources |
| Embedding tokens | 5M/mo per model | 10M/mo per model | Unlimited (rate-limits table) |
| At limits | Requests fail with 403 or 429 errors | Blocked, not billed | Metered pay-as-you-go |
The Bottom Line
Pinecone’s Starter plan is a real free tier: $0, no monthly minimum, serverless billing, and monthly embedding and rerank allowances. Three limits are the ones to plan around: the us-east-1 and AWS-only lock-in, the 1 GB vs 2 GB storage figures that Pinecone’s own pages give differently, and the hard request failures once object or rate limits are reached. Builder at $20/month flat is the natural next step for most indie hackers because it adds other clouds, other regions, and a much larger token allowance without overage billing.
FAQ
Is Pinecone’s free tier really free, or will I get a surprise bill?
Starter is $0/month with no monthly minimum, according to the Pinecone pricing page. Builder is a flat $20/month with no usage overages, and Standard has a $50/month minimum before pay-as-you-go. When you exceed an object limit, the API returns a 403 QUOTA_EXCEEDED error, and when you exceed a monthly rate limit it returns a 429 error, per the rate-limits reference.
Can I use Pinecone’s free tier outside us-east-1 or on Azure or GCP?
No. Starter runs only on AWS in us-east-1, according to the pricing page and the object-limits reference. Builder at $20/month is the first plan that allows other clouds and regions. If your users are far from the U.S. East Coast, plan that upgrade early.
What happens if I exceed the 1 GB storage or 500k tokens on Starter?
The pages do not say whether exceeding storage blocks writes, blocks reads, or pauses the index, so treat that behavior as unverified. Monthly read-unit, write-unit, and embedding-token limits return a 429 TOO_MANY_REQUESTS error until you upgrade, per the rate-limits reference. The free-tier inactivity policy is also not stated in the pages reviewed here.
How This Guide Was Built
This guide was written on October 14, 2026 from the Pinecone pricing page, the understanding-cost docs, the database limits overview, the object-limits reference, the rate-limits reference, the save-on-costs guide, the pod-based indexes guide, the upgrade guide, the dedicated read nodes overview, the quickstart. We did not create a free account, load any documents, run queries, or measure latency. The two Pinecone pages give different storage figures for both Starter and Builder; we report both rather than choose one.