Usage Based vs Subscription Pricing: Retrieval Break-Even Guide
Usage based vs subscription pricing depends on workload and billable units, with a break-even guide to metered, fixed, and hybrid retrieval plans.
A procurement or platform lead comparing retrieval prices needs the same monthly workload and billable-unit definitions for each offer. Calculate the break-even point after included usage, storage, writes, model costs, and retained operating work are accounted for. This guide focuses on that calculation rather than a general catalog of pricing models.
Key takeaways
Choose a retrieval pricing model by matching its billable units to your monthly workload.
- Use metered pricing when demand varies and you can measure reads, writes, storage, and other usage.
- Use a monthly fee when stable demand fits the capacity and services included in the fee.
- Use a hybrid plan when a predictable baseline and variable usage need to share one budget.
- Calculate break-even with the same workload, unit definitions, and cost boundary for each option.
How metered, subscription, and hybrid pricing differ
A billable unit is the provider-measured quantity that drives a usage charge. A provider may measure queries, written data volume, or retained storage.
A monthly minimum is a floor applied according to the contract, not automatically an added base fee. Where it applies to eligible metered charges, calculate the amount as the greater of the minimum or those charges, then add only charges outside that floor. Two vendors can both advertise usage-based pricing while charging for different units, so the model label alone does not predict the invoice.
| Model | How the bill is formed | Budget pattern | Typical fit |
|---|---|---|---|
| Metered usage | The provider charges for measured operations or resources. The rate can vary by reads, writes, storage, or another unit. | The bill moves with consumption. A low-usage month can cost less, while a usage surge can raise the bill. | A pilot, seasonal service, variable traffic, or a workload that the team is still measuring. |
| Fixed monthly subscription | The buyer pays a recurring fee for stated capacity or a defined package. | The recurring fee is easier to forecast when usage stays inside the included amount. | A stable workload whose normal and peak months fit the included service. |
| Hybrid | The offer may combine a recurring fee with an allowance and overages, or impose a minimum-spend floor on eligible metered charges. These are different billing rules; apply the contract terms. | A recurring fee can provide a predictable baseline; usage-based charges or a minimum floor depend on the eligible units and actual terms. | A steady baseline with bursts, or a service that combines a recurring package with metered dimensions, storage, or other use. |
In a pure metered model, the cost per month can change even when the provider has no monthly subscription fee. Storage held for more days, a larger write volume, and a higher number of retrieval operations can all change the amount. A buyer should therefore translate expected work into the provider’s units before comparing the monthly total.
Chroma Cloud charges separately for reads, writes, and storage, per its pricing documentation; a write through add, update, or upsert costs $2.50 per logical GiB, the raw, uncompressed data sent to the service. Storage is measured in GiB-hours, which combines the amount stored with the time it remains stored, and billing aggregates collections and tenants within one Chroma Cloud account. Include write volume and retained storage alongside reads when estimating a Chroma Cloud workload.
For a minimum-spend offer, identify which usage charges count toward the floor and which resources are billed separately. Do not model the floor as an automatic base fee added to every metered charge; follow the offer’s current contract and pricing rules.
Under Weaviate’s billing documentation, Shared Cloud clusters require a billing account, and charges continue for each active cluster until it is deleted. Include each cluster’s active period, from test creation through deletion or production operation, in the cost model.
Read the unit before comparing the rate
A retrieval request is not a universal billable unit. A request can trigger a read or query, while retrieval services may also bill for writes, retained storage, or stored vector dimensions.
Chroma bills once per write, with no additional write charge for background compaction or reindexing, per its pricing documentation.
Weaviate vector-dimension rates vary by cloud provider and region; its pricing details display “from” rates for the cheapest optimization profile. Vector-dimension prices also vary by index type and compression, per the Weaviate product pricing page. For an estimate, record the intended region, index type, compression choice, stored dimensions, and storage requirement so the price reflects the planned configuration.
Weaviate lists core database features, role-based access control, and automated upgrades for Flex. Its AI services are available on the plan but have their own billing: Embeddings are usage-based, while Query Agent has a free tier and usage-based charges beyond it. Model these costs separately from the database minimum, and compare each offer’s actual operating responsibilities and labor.
How to calculate a workload break-even point
A break-even calculation compares the cost of the same monthly work under each plan. Start with the service boundary, map workload to each provider’s billable units, and compare low, expected, and high months. Then add costs that sit outside the retrieval line item, such as embedding generation, model calls, network transfer, backups, or in-house operations when they apply.
1. Define equivalent work
Describe the workload in operational terms before opening a price sheet. Quantify monthly reads, documents added or updated, average and peak retained data, retention time, and the vector or index configuration. Include how many tenants or services share the account when billing aggregates those resources.
For an illustrative workload, assume a support assistant serves 80,000 retrieval requests in a month from a mostly stable index, while a catalog ingestion run writes many new product records before traffic begins. Assume a separate knowledge assistant has fewer requests but keeps several years of source material online.
The FinOps Foundation’s Unit Economics capability separates resource-efficiency measures, such as cost per GB stored or transferred, from business measures such as cost per transaction, tenant, or resolved case. Use both views for a retrieval service: cost per million billable reads helps compare resource use, while cost per supported request or successful business task tells the organization whether the spend serves its purpose.
2. Build the monthly cost equation
For a simple usage model, calculate each unit separately:
Metered monthly cost = (read units × read rate) + (write units × write rate) + (storage units × storage rate) + other metered charges.
Hybrid monthly cost = base fee + charges for usage outside included amounts + separately metered resources.
For a flat subscription, use the recurring charge plus any costs that the package leaves outside its scope. The same data, retention, read pattern, and supporting services belong in every candidate’s calculation. If one plan includes an embedding service while another requires a separate model account, include the model cost or the included-service value in the same comparison boundary.
Break-even is the monthly workload at which the two calculated totals are equal. For a simple comparison between a metered plan at rate p per unit and a flat monthly fee F, the equation is p × workload = F. Thus, the break-even workload is F ÷ p. Below that point, the metered plan has a lower calculated charge; above it, the flat fee has a lower calculated charge, provided the workload stays within the flat plan’s included scope.
For a package with an included allowance, enter monthly fee F, included units M, and overage rate r: cost is F through M, plus r for each unit above M. Compare that total with the metered alternative’s p × workload.
When the break-even point falls above M and the metered rate p exceeds overage rate r, solve for workload = (F − rM) ÷ (p − r). Below the allowance, compare F with p × workload, and check the candidate intersection against workload ≤ M. Above the allowance, use the stated formula only when its result is greater than M; when p ≤ r, inspect each range directly. For several units or price tiers, calculate each range separately.
A minimum-spend floor is different: where the minimum applies to eligible charges, use the greater of the minimum or those charges, then add charges excluded from the floor, according to the contract.
Compare the same cost categories and boundary for each offer, but use each offer’s actual model, labor, and other costs rather than assuming those amounts are equal.
3. Work through low, expected, and peak months
The following hypothetical calculation uses assumed—not vendor-published—prices and volumes for the same retrieval behavior, with other costs excluded. Assume a metered service charges $0.002 per retrieval, while a hybrid service costs $240 per month including 100,000 retrievals, then $0.0015 for each additional retrieval.
| Monthly retrievals | Metered plan at $0.002 each | Hybrid plan at $240, including 100,000, then $0.0015 each | Lower modeled charge |
|---|---|---|---|
| 60,000 | $120 | $240 | Metered |
| 150,000 | $300 | $315 | Metered |
| 180,000 | $360 | $360 | Equal |
| 240,000 | $480 | $450 | Hybrid |
At low forecast volumes, the recurring portion may outweigh consumption charges, so account for package functions that would replace separate tools in your design. At high forecast volumes, the example’s lower overage rate drives the hybrid plan’s modeled advantage, making the post-allowance rate the key term to evaluate.
The FinOps Foundation’s cloud forecasting playbook recommends defining the purpose of a forecast, such as aligning a budget or managing commitments, because that choice sets the needed level of detail. It also recommends forecasts based on future plans as well as historical usage. For a retrieval pilot, add the planned launch date, customer or document growth, retention changes, and scheduled reindexing to the forecast instead of extending last month’s request count unchanged.
4. Use the published Seahorse inputs carefully
The Seahorse Cloud price sheet, checked on October 19, 2026, lists the separate API rates as KRW 16,000 per 1M query RUs, KRW 6,000 per 1M write WUs, and KRW 50 per GB of storage per month.
At those listed rates, 2M query RUs equal KRW 32,000, 3M write WUs equal KRW 18,000, and 100 GB stored for one month equals KRW 5,000 in storage-rate arithmetic. These rate-only calculations use the published rates; they are not a Standard invoice. Apply the plan’s included amounts, minimum or base-charge rules, and credit terms before calculating the payable total. Confirm how the combined RU/WU allowance is allocated. A retrieval request is not automatically one RU; map the actual operation to the provider's metering definition.
5. Make the break-even repeatable
Use a small spreadsheet or cost model with one row per billable unit and one column per candidate. Add a separate column for each scenario month. The model should make it possible to change request volume, new writes, retained storage, or a plan allowance without rebuilding the calculation.
What published retrieval-service plans show
Vendor prices in this guide were checked on October 19, 2026. The plans use different currencies and billable units, so the table compares pricing structures rather than ranking price.
| Service and plan | Published base or entry term | Published metered or included resources | Billing detail to carry into a cost model |
|---|---|---|---|
| Chroma Cloud | Usage-based pricing | Reads, writes, and storage; writes cost $2.50 per logical GiB sent through add, update, or upsert. | Storage is measured in GiB-hours, and billing aggregates all collections and tenants in the account. A collection fork costs $0.03 per fork request. |
| Weaviate Cloud Flex | Starts at $45 per month; the database minimum covers baseline vector-dimension and storage charges | Pay-as-you-go; Embeddings are usage-based, and Query Agent has a free tier plus usage-based charges. | Rates vary by region, index type, and compression; AI-service charges are separate from the database minimum. |
| Seahorse Cloud Starter | Usage-based prepaid billing; one-time KRW 50,000 usage credit for one month | Metered service use | Treat the credit as a one-month pilot input, not a recurring reduction. |
| Seahorse Cloud Standard | Starts at KRW 30,000 per month | Usage-based postpaid billing; 1M Seahorse API RU/WUs included | Query RUs, write WUs, and storage have separate published rates, as detailed in the break-even example. |
Chroma Cloud is serverless, so customers do not provision servers for its operation, per Chroma Cloud documentation. That operating model is useful for an early workload estimate because the buyer can focus the price comparison on measured service use rather than a self-managed server’s provisioned capacity. The whole-service cost still includes the team’s integration, data preparation, and model work when those remain on the buyer’s side.
Chroma applies input and query quotas across user-facing operations, per its Quotas & Limits documentation. Before setting production assumptions, obtain the quota values for the intended account and compare them with pilot batch sizes and query patterns.
When each pricing model fits
| Workload or budgeting need | Model to evaluate first | Why it can fit | Example plan fact |
|---|---|---|---|
| A pilot with uncertain request and ingestion volumes | Metered usage or prepaid usage | Track measured use while prepaid terms can support a time-limited trial budget. | Chroma Cloud pricing covers reads, writes, and storage; compare the units with the pilot forecast. |
| A steady service with consistent consumption | Fixed monthly package | A recurring amount is easier to plan when the expected workload fits the included capacity and service scope. | Compare the full allowance and required functions with the service’s expected normal and peak month. |
| A stable baseline with short peaks | Hybrid or usage-based pricing with a defined minimum | A recurring package or a correctly modeled minimum floor can cover baseline costs while metered charges follow variable consumption. | Compare each offer’s eligible units, included amounts, and minimum-spend rules with the forecast. |
| Seasonal or campaign-driven retrieval demand | Metered or hybrid pricing | A workload forecast can represent quiet and busy periods separately, rather than treating peak traffic as the monthly average. | Compare a low, expected, and peak month against each vendor’s own units and minimums. |
| A budget owner who needs a stable monthly commitment | Fixed package or hybrid with a capped, contractually defined variable amount | The recurring part supports budget planning; an agreed variable limit gives finance a known decision point. | Record the commitment, included resources, extra-use rates, and any alert or cap terms in the service order. |
A low-traffic month favors metered usage when a recurring fee would exceed the actual measured bill and the service still meets its operational needs. In a high-traffic month, the same per-unit charge can grow beyond a subscription fee, while the subscription remains a better fit only if it covers the required volume and avoids material extra charges.
Cost and control checks before a purchase
Assign an owner to keep the production estimate current as workloads and plan terms evolve.
- Count reads by the provider’s unit. Record the billable read definition, scanned data, and result volume. Chroma charges $0.0075 per TiB queried and $0.09 per GiB returned, per its read-pricing documentation.
- Estimate writes and refreshes. Record new and changed records, reprocessing, and data volume; separate initial loads from recurring refreshes when their volumes differ.
- Measure retained data over time. Estimate storage and retention for normal and peak months. Chroma measures storage using GiB-hours, the stored data amount accumulated over time. For any minimum-spend plan, verify which storage and vector-dimension charges count toward the floor and which resources are billed separately.
- Include related model and network costs. Add separately billed embeddings, answer generation, reranking, data transfer, and backups.
- Price the support and operating boundary. Include support, ingestion, monitoring, upgrades, and engineering work left to your team; account for managed functions included in the package.
- Test low, expected, and peak months. Replay representative pilot traffic and ingestion, using the same workload, dimensions, retention, and service boundary for each candidate.
- Set a budget owner and alert path. Name the alert recipient, review threshold, and follow-up action. Use available provider controls and assign an owner to pause or remove test resources after the pilot.
- Reconcile the forecast with the invoice. Compare actual units and charges with the estimate monthly. The FinOps Foundation’s cloud forecasting playbook recommends using expected future workloads as well as historical usage.
An alert is most useful when it points to a choice the team can make. For example, a platform owner can review query patterns when read units rise faster than completed requests, or review the retention policy when stored data rises while active knowledge sources remain stable. A product owner can approve higher spend when it follows planned adoption and an agreed service outcome. These actions connect a bill to the workload rather than treating every increase as an error.
FAQ
What is a usage-based pricing model?
A usage-based pricing model charges for measured consumption, such as reads, writes, or storage. The monthly total changes with the provider’s billable units, so estimate the workload in those units before comparing it with a recurring plan.
Is a monthly minimum the same as a fixed subscription?
A monthly minimum sets a floor on eligible usage-linked charges, while a fixed subscription charges a recurring fee for a defined package or allowance. For a minimum-spend plan, apply the floor to the charges specified in the contract; do not automatically add the minimum to all metered usage.
Can I compare retrieval providers by cost per query?
Use cost per query as one measure, then add writes, retained storage, and other billed resources. Chroma Cloud, for example, prices reads, writes, and storage as distinct usage dimensions, and Seahorse Cloud publishes separate query RU, write WU, and storage rates.
When should a team revisit its break-even estimate?
Recalculate when traffic, document volume, retention, index configuration, model use, or plan terms change. Compare forecast amounts with actual monthly units and invoices so that a price-model decision continues to reflect the workload in production.
Choosing a plan with a repeatable cost boundary
Choose metered usage when demand is variable and billable units are measurable; a fixed package when steady demand fits its included scope; or a hybrid plan when baseline and variable charges remain predictable across scenarios. Compare the same workload, period, cost categories, and service boundary, but calculate each offer’s actual model and labor costs and apply included usage, minimums, and overages under its terms. Validate the choice against low, expected, and peak months and the first production invoice.
Related background: Estimating Retrieval Capacity from Storage and Query Demand
Try Seahorse Cloud
Evaluate this workflow with your own documents and acceptance criteria.
