Send every request to parezen/auto. Parezen picks the cheapest model that meets your quality bar and proves every saving against your provider bill. No subscription: 8% on credits, nothing on your own keys.
Pricing updated 4 Nov 2026. Free, read-only, no card. Sign up in a minute.
No subscription. Your own keys are free to $10K a month. No card to start.
Read-only to start. No change to your traffic.
Every feature on every plan.
Two plans, published fees
Every feature is on every plan. Pay as you go has no subscription: 8% when you add Parezen credit (at least $1 by card), with models at the provider's list price. Your own provider keys: free to $10K a month, then 3%. Enterprise adds controls, invoicing and an SLA under an annual contract.
Parezen works with OpenAI, Anthropic, Google, Mistral and more, directly or through AWS Bedrock, Azure OpenAI and Google Vertex. Hosting is included: there is no separate hosting fee.
What Parezen charges for
What
Price
Notes
Models on Parezen credits
8% when you add credit
At least $1 by card. Models at the provider's list price for the model served, no markup.
Models on your own keys
Free to $10K a month, then 3%
Your providers bill you directly. On traffic we optimise, the fee is never more than the savings. Charged monthly once it reaches $25.
Auto routing, quality checks and receipts
Included
Quality checks are normal usage on your account.
Savings share
15% of verified savings
Optional, for 12 months per change, capped at 5% of pre-change spend.
Enterprise controls and SLA
Annual contract
Usage fee on the invoice, which a contract may discount.
Pay as you go
8% on credits
Every feature, no subscription. Your own provider keys are free to $10K a month.
Auto routing with your quality bar, and every saving on a receipt
Parezen credits: one key for every model, 8% when you add credit
Your own provider keys: free to $10K a month, then 3%, never more than we save you on the traffic we optimise
15% of verified savings, for 12 months per change, capped at 5% of the pre-change AI spend of the workloads we optimise (measured in your savings forecast). Optional, for any team with traffic through Parezen. Only savings forecasted against your provider bill count.
Free
Savings forecast
Savings forecast: a read-only look at your AI bill, from an uploaded export or a read-only billing connector. No card.
$5,000
Verified Savings Assessment
Optional $5,000 Verified Savings Assessment, credited. It takes 2 weeks, is verified against the invoice, and is credited in full against a contract signed within 60 days. Open to every customer.
Two ways to pay for models
Keep your provider accounts, or use one Parezen key for every model and pay from a prepaid balance. Every feature works with both, and you can switch.
Bring your own keys
Free to $10K a month, then 3%
A pz_ key plus your provider keys, stored encrypted. Your providers bill you directly.
Add a provider key and every model from that provider is ready to use.
Only the optional savings share, if you choose it.
Billing never blocks traffic on your own keys.
Parezen credits
8% when you add credit
One pz_ key. Every model works, with no provider accounts. Parezen pays the providers.
Usage at the provider's list price for the model served, with no markup. The 8% fee is added when you buy credit: a $100 top-up costs $108.
Card top-ups from $20 (at least $1 fee).
$5 starter credit. Warnings at 20% and 5% left, and auto top-up keeps you running.
How Parezen runs
Parezen runs the gateway for you at api.parezen.com. You change one base URL and use a pz_ key, and there is nothing to install. Hosting is included. Or run it in your own cloud with a self-hosted licence.
Hosted gateway
Hosted
Change one base URL to api.parezen.com and use a pz_ key. Nothing to run.
Bring your own provider keys. Parezen stores them encrypted and accepts them only while encryption in transit, encrypted storage and rate limiting are all on.
No separate hosting fee and no trial limit: you pay only the usage fee on credits.
Metadata only by default. Request logs kept about 30 days.
Self-hosted
Enterprise
The same gateway on your Kubernetes or Docker, with your Postgres. Prompts and provider keys never leave your network.
$30K a year licence for 5 gateways, plus 15% of savings.
You pin the provider hosts it may call; no config update from Parezen can change them.
If Parezen is unreachable
In your app
Your app can send the call straight to your provider when Parezen cannot be reached, times out before answering, or returns a 5xx.
Never on a 4xx: a budget hard stop must stay a stop.
Calls sent direct are not routed, recorded or billed.
Cloud models
Included
Models you already buy on Bedrock, Azure OpenAI or Vertex, through the same gateway.
AWS and Azure Marketplace private offers come after launch. Until then, buy direct.
The savings share: pay only for savings we can prove
The figures below are pricing examples. They are not savings measured on a customer bill.
The savings share is optional: 15% of verified savings, for 12 months per change, capped at 5% of the pre-change AI spend of the workloads we optimise (measured in your savings forecast). If the savings, after the cost of quality checks, do not exceed the fee, the share costs nothing that month. A saving is verified only when the month's ledger matches your provider bill.
If we don't save you more than we charge, you pay nothing that month. That sentence is the monthly savings share only, judged on savings after the cost of quality checks. It does not apply to the usage fee.
Illustrative example
A request used to cost $1.00. After Parezen's changes it costs $0.40, so you save $0.60. Our share is 15% of the saving ($0.09), but the fee can never be more than 5% of the pre-change spend: 5% of $1.00 is $0.05. So you pay $0.05 and keep $0.55 of the saving.
The cap only matters once savings pass about 33% of the original cost.
Illustrative example
Pre-change AI spend
Verified savings
15% share
5% cap (of $40,000)
You pay
$40,000
$8,000 (20%)
$1,200
$2,000
$1,200
$40,000
$16,000 (40%)
$2,400
$2,000
$2,000
How we calculate savings
These are the rules the ledger applies. You can export the ledger and price book and recompute every number.
Your baseline is your usual model: the model each key or team names as its default. Auto routing saves against it, and the saving is the usual model's cost minus the actual cost, at the same price version.
For the savings share, the baseline is the cheaper of your usual model and the model you actually used most in your first 30 days through Parezen. Those 30 days are the pre-change spend for the cap, and earn no share.
The baseline is re-priced at today's list prices. A provider price cut goes to you, and does not count as a Parezen saving.
Only a change Parezen makes counts, each for 12 months. Savings you already had, and anything you change yourself, are excluded.
A cheaper model is used only after it passes your quality bar, and is checked on a sample of live answers. If quality drops, it is removed and the savings since its last passing check are not counted or charged.
Each month the ledger is reconciled with your provider's bill, and the more conservative figure is billed.
The formula on this page is published. Your ledger and price book are open to export, so you can recompute every number.
Enterprise is an annual contract: SSO and SCIM, model and provider allowlists, zero data retention, EU or US region, invoiced billing at provider prices with a contract usage fee, and an SLA. Talk to us for terms. AWS and Azure Marketplace private offers come after launch.
There is no subscription. 8% when you add Parezen credit (at least $1 by card), with models at the provider's list price. Your own provider keys: free to $10K a month, then 3%. The savings share is optional. Enterprise is an annual contract.
Models are charged at the provider's list price, with no per-token markup. On Parezen credits, the 8% fee is added when you buy credit: a $100 top-up costs $108 and adds $100. Quality checks are normal usage, so they carry the fee too.
No. The savings forecast needs no card, and new organisations get $5 of starter credit.
Billing never errors traffic on your own keys. On credits, a zero balance returns 402 with a top-up link, after warnings at 20% and 5% left; auto top-up prevents it. Only the budget hard stops you set yourself also return an error. If Parezen itself is unreachable, your app can send the call straight to your provider; the fail-open guide shows how.
Send parezen/auto and Parezen uses the cheapest model your team has approved. A model is approved only after it matches your usual model on your test questions and on live checks, and it is removed at once if it drops below your quality bar. Every response carries its cost, its saving and why that model was chosen.
15% of verified savings for 12 months after each change, capped at 5% of the pre-change AI spend of the workloads we optimise (measured in your savings forecast). Savings are your baseline cost, re-priced at today's list prices, minus your actual cost. Only a change Parezen makes counts, and each month the ledger is checked against your provider invoice; the more conservative figure is billed.
No. Your baseline is re-priced at today's list prices, so price cuts go to you, not into our fee.
No. It is optional, and open to any team with traffic through Parezen. Every feature works without it.
Enterprise is an annual contract: SSO and SCIM, model and provider allowlists, zero data retention, EU or US region, invoiced billing at provider prices with a contract usage fee, and an SLA. Talk to us for terms. AWS and Azure Marketplace private offers come after launch.
AWS and Azure Marketplace private offers come after launch. Until then, buy direct.
The savings forecast is a read-only look at your AI bill, from an uploaded export or a read-only billing connector, with no card. Optional $5,000 Verified Savings Assessment, credited. It takes 2 weeks, is verified against the invoice, and is credited in full against a contract signed within 60 days. Open to every customer.
No. Upload usage exports from LiteLLM, OpenRouter, Portkey or your provider bills (Excel, CSV or JSONL with a date, a model and token counts) and Parezen shows where the money goes. To cut spend and verify savings, route through the Parezen gateway.