Simple, Transparent Pricing
What you send us decides which model governs your bill.
Send raw files. Mixpeek handles extraction, embedding, and indexing. Pay for what your agents extract. Searches and retrievals come out of your plan's included pool.
Shared multi-tenant infrastructure. Pay per unit of content processed.
Build
Processing pool included in the minimum, then metered per unit of content. Up to 100K objects/month.
- Basic extractors
- Webhooks
- Batch processing
- Role-based access control
- Email support
- All extractors
- API key management
Scale
Processing pool included in the minimum, then metered per unit of content. Up to 1M objects/month, SSO, and priority support.
- Basic extractors
- Webhooks
- Batch processing
- Role-based access control
- All extractors
- API key management
- SSO integration
Enterprise
Single-tenant infrastructure, BYO extractors, SSO, audit logs, and a dedicated support team.
- Dedicated deployment
- SSO, audit logs & compliance review
- Forward-deployed engineer
- Dedicated Slack support channel
- Custom integrations
- Custom plugins & extractors
- Custom model training & fine-tuning
- Dedicated support team & SLA
How Mixpeek compares
Most vector stores are an index. MVS is a database with agent-native retrieval: so you stop paying to bolt those capabilities on yourself.
| Capability | Pinecone | Qdrant | S3 Vectors | MVS |
|---|---|---|---|---|
| Database operations on your vectors | ||||
| Dense vector search (ANN)Approximate nearest neighbor search over high-dimensional embeddings. The baseline capability every vector store needs. | ||||
| Native dense + sparse + BM25 hybrid searchFuse semantic, sparse, and keyword relevance in one query plan. S3 Vectors is an index; MVS runs hybrid retrieval natively. | Dense + sparse | Dense + sparse | Dense + sparse + BM25 | |
| Semantic JOINs across namespacesJoin two namespaces by vector similarity, like a SQL JOIN but on embeddings. No denormalization or data duplication required. | ||||
| Aggregations (GROUP BY, COUNT, SUM, AVG)Run analytics directly on your vector store. Group documents by metadata fields and compute counts, averages, and sums without ETL. | ||||
| Cross-shard transactionsAtomic writes across multiple shards using two-phase commit, so multi-namespace changes remain all-or-nothing. | ||||
| Time-travel queriesQuery data as it existed at a past point in time by replaying the write-ahead log. Useful for debugging, auditing, and reproducibility. | ||||
| Object storage-native persistenceData lives in your object storage. S3 Vectors gives you a cheap index; MVS adds database operations on the same storage foundation. | ||||
| Built for agentic workloads | ||||
| Streaming partial resultsGet results as shards respond instead of waiting for every shard. Agents can evaluate early hits and decide whether to refine or cancel. | ||||
| Query cancellationCancel in-flight fan-out queries that an agent no longer needs. Freed shard work returns to the pool instead of burning through the loop. | ||||
| Per-agent budget limitsEnforce max queries, writes, and compute per agent or API key at the coordinator level. Prevent runaway autonomous loops from running up spend. | ||||
| Standing queriesRegister a persistent query that fires a webhook whenever a newly ingested document matches. | ||||
| Multi-stage retrieval pipelinesChain retrieval stages such as broad recall, filtering, joining, and reranking in one request instead of wiring round trips in application code. | ||||
| Query audit logFull audit trail of every query: who ran it, when, and what was returned. Agents need inspectable retrieval, not black-box calls. | ||||
Enterprise
Dedicated infrastructure for production AI retrieval
Pricing
From $10,000 /month
A platform license plus managed infrastructure, priced on your deployment model, support tier, and scale. We'll scope it with you.
Working to a fixed budget or a quarterly cycle? We also scope fixed-price engagements against a defined outcome, so you can approve one number instead of forecasting usage.
Deployment options
- Mixpeek-managed dedicated tenant
- Customer-managed cloud (AWS / GCP / Azure)
- Hybrid networking
What's Included
- Dedicated deployment
- SSO, audit logs & compliance review
- Forward-deployed engineer
- Dedicated Slack support channel
- Custom integrations
- Custom plugins & extractors
- Custom model training & fine-tuning
- Dedicated support team & SLA
- Unlimited usage & collections
Frequently Asked Questions
What are Mixpeek's current usage rates?
As of August 2026: video $0.05/minute and audio $0.01/minute processed, images and document pages $1.50 per thousand, text $2 per million tokens, web pages $20 per thousand crawled. Vector storage is $1.56 per million vectors per month. Which line dominates your bill depends on your mix: storage for a large corpus you keep and search, ingestion for a high-volume crawl or a big video library. Queries are $2 per million beyond your plan's included pool. Plans start at $25/month (Build, $10 usage pool) and $250/month (Scale, $100 pool). The interactive calculator above always reflects live rates.
Does the per-minute price include the AI model cost, like Gemini?
Yes. A feature's rate is what you pay in total for that feature; there is no separate model bill. The rate card has two kinds of row and the difference is where the cost comes from. Fleet rows run on our own GPUs at a fixed sticker: video search $0.05/minute, faces $0.10/minute, on-screen text $0.10/minute. External rows are cost-plus on a third-party model, so the model's cost is inside the number you see: multimodal understanding is $0.0292/minute of video, $0.50 per 1,000 images, $2.00 per 1,000 document pages. You are never invoiced separately for Gemini, OpenAI or anyone else.
Does my price change if I use a cheaper model, or bring my own?
For external rows, yes, because they are priced cost-plus rather than at a fixed sticker: when the underlying model's cost moves, the rate moves with it. Fleet rows do not change, since they run on our hardware at a fixed price regardless of what model is configured. Bringing your own weights is a single-tenant arrangement rather than a shared-tenant one, for a practical reason: custom weights need dedicated GPU memory and a deployment that can be restarted around your model, which is not something a shared fleet can do without affecting other tenants. On single tenant you pay a platform fee plus the compute your deployment actually uses, so a cheaper or smaller model shows up directly as lower compute.
What does "priced by features extracted" actually mean?
Each thing you want to search by is a separate line on the rate card, priced in that file type's natural unit, and you pay only for the ones you enable. A base rate makes the content searchable at all (video $0.05/minute, images and document pages $1.50 per thousand). Each additional feature adds its own per-unit rate on the same meter: faces, on-screen text, audio fingerprint, multimodal understanding. Clustering and taxonomy enrichment are included at no extra charge. So indexing an hour of video for scene search alone is $3.00, and the same hour with faces and on-screen text is $15.00. Adding multimodal understanding at $0.0292/minute puts that hour at $16.75. Nothing is bundled into a number you cannot break down.
Does the language of my content change the price, or what works?
The price does not change. Video is billed per minute and documents per page whatever language the content is in, and there is no surcharge for non-English material. Quality does vary by feature, because each one inherits the model underneath it. Embeddings, descriptions and transcription on the multimodal rows run on Gemini, so they cover what that model covers. On-screen text is read by a vision model, so script coverage follows that model and non-Latin scripts are worth testing before you commit. Face detection and audio fingerprinting never read text, so they behave the same in any language. If your library is not in English, bring a sample to a workshop session and see the output on your own content before you sign anything.
What does speech transcription cost?
There is no separate transcription line on the rate card. Transcription runs as a step inside multimodal understanding, which handles the speech alongside the rest of the file, so you pay that feature's rate for the file type: $0.0083 per minute of audio, $0.0292 per minute of video. Enabling multimodal understanding is what gets you transcripts, and nothing on your invoice will read "transcription". You cannot buy transcription on its own today.
Can my company pay by invoice instead of a card?
Yes, ask us. The self-serve path is a card through Stripe because that is what gets a workspace open in a few minutes, but it is not the only path. If your organization cannot expense a card, or needs the bill to arrive against a purchase order or a team account rather than an individual, email [email protected] and we will set that up; there is a direct link for it on the plan page during signup. Single-tenant Enterprise is a contract and a provisioned environment rather than a signup, so it is invoiced by default. What we do not do is make you find this out only after you have created an account.
What is the difference between indexing cost and storage cost?
They are different bills and the first one is usually much larger. Indexing is one-time work per item: running the extractors over your content at the rates above. Storage is what you pay every month to keep the results queryable: $1.56 per million vectors per month, $0.023/GB/month for object storage, and $25/GB/month for the hot cache tier that keeps a working set in RAM. The per-vector term is the dominant part of the monthly figure. A 1.2M-vector index is about $1.90 a month to keep, which says nothing about what it cost to build: indexing the library that produced those vectors is the number to plan around. Re-indexing, for example after changing an embedding model, means paying the indexing side again.
Who pays the cloud bill on a single-tenant deployment?
It depends on whose account the infrastructure runs in, and both arrangements exist. When Mixpeek hosts your single-tenant deployment, you pay a platform fee plus the cloud cost of that deployment, billed through us at a 20% handling fee on top of the underlying cloud cost, so it arrives as one invoice. When you host it in your own cloud account, you pay your provider directly for compute and storage, we never see that bill, and you pay us the platform fee alone. The choice is usually about procurement and data residency rather than total cost, and the honest way to compare them is to price your workload both ways before committing.
Shared tenant or single tenant: what actually changes?
Shared tenant is the self-serve path and is the right default: your data is isolated by namespace, you are on the published rate card, you start at $25/month, and you can change your mind cheaply. Single tenant gives you dedicated infrastructure, and the things that genuinely require it are custom model weights, a specific cloud region or account, and compliance rules that will not accept shared infrastructure. What you gain is control and a support relationship with committed response times. What you give up is the ability to start in an afternoon, since single tenant is a contract and a provisioned environment rather than a signup. If you are unsure, start shared: moving to single tenant later is a migration we do regularly, and provisioning single tenant for a workload that never needed it is money spent on nothing.
What's the difference between Standalone and Managed?
Standalone (MVS) is a pure vector store: you bring your own embeddings and pay for object storage, hot cache, queries, and writes. Managed includes Mixpeek's extraction pipeline: you send raw files (video, images, PDFs), and Mixpeek handles chunking, embedding, and indexing, billed per unit of content processed.
How is Managed usage billed?
Three questions set your bill: what kind of files (video, image, audio, document, text, web), how much of them (each file type is billed in its honest natural unit: video and audio per minute, images per file, documents per page, text per token), and what you want to search by (visual similarity, faces, on-screen text, multimodal understanding -- each adds a per-unit rate on the same meter; clustering and taxonomy enrichment are included at no charge). Your plan's monthly minimum includes a usage pool that absorbs the first dollars of processing; you only pay overage beyond it. Searches and retrievals are included.
Can I use MVS as just a vector store without extraction?
Yes. MVS works standalone. Bring your own embeddings, upsert directly, and search with dense, sparse, or BM25 hybrid queries. Build starts at $25/mo with up to 1M vectors. You can add extraction pipelines later without rebuilding your retrieval layer.
How does enterprise pricing work?
Enterprise is a custom platform license plus managed infrastructure, priced on your deployment model, support tier, and scale. Deploy as a Mixpeek-managed dedicated tenant, in your own cloud (AWS/GCP/Azure), or hybrid: and bring your own extractors and models. Contact sales for a quote.
Are there volume discounts?
Yes. On Build and Scale, usage beyond your minimum is billed at standard rates with volume discounts at higher tiers. Enterprise customers get custom pricing based on committed usage, deployment model, and support requirements.
Are there any long-term commitments?
No. All self-serve plans are billed monthly with no long-term commitments. You can upgrade, downgrade, or cancel at any time.