Skip to Content

Models and billing

Models, what they cost, and how you pay.

Models are what agents, Code, API keys and chat call, each by its alias. Each call has a price. Where the model comes from decides who pays it.

Use it when
You pick a model, or want to know what its calls cost
You get
OpenOrange's models, your own provider keys, and routing pools
Cost
The model's listed price; nothing on your own key
Setup time
One form per provider key you add
Needs
See models; every member has it

Where models come from

Models lists every model your Sunport offers. Most come from one of three sources:

  • OpenOrange: models OpenOrange provides. Each call is charged to your balance.
  • Your own provider key: a model on a key you added under Manage providers. Your provider bills you; OpenOrange charges nothing.
  • Routing pool: one alias that picks among models you choose, for each new conversation.

Routing pools also lists Built-in routers. Each picks a model for every request.

What a call costs

A model's page shows its price under Price and billing:

  • Text models show a price Per 1M tokens, input and output. Cached input is the price of input read from the prompt cache.
  • Some models charge per image, per second of video, per audio hour or per 1K characters.
  • Some prices change in Peak hours, shown in UTC.

How it's billed says who charges for the call:

  • An OpenOrange model: OpenOrange charges each call to your balance at these prices.
  • Your own provider key: your provider bills you, and OpenOrange charges nothing.
  • A routing pool: the price of the model it picks, plus a small charge for picking. That charge shows on its own as Auto routing.
  • A built-in router by a supplier: an estimate first, then the final cost from the supplier's receipt. The final cost can be less or more.
  • A video: charged when it starts, for the seconds it asks for. The model's page says when a failed video is refunded.

Each call's cost shows on Requests. Each model's use and spend shows on Models.

Who can use a model

Every member can see the models they may use, with their prices and their own use. Only administrators see everyone's use and spend.

By default, everyone may call every model, including ones added later. An administrator can narrow that:

  • For a person, under Limits on their page.
  • For an agent, under Budget on its Models tab.
  • For a key, under Allowed models on its page.

A call to a model outside the list is refused.

Badges on a model

A badge on Models explains a refusal or a choice:

  • Not in trial: a free trial leaves out the most expensive models. An agent set to one answers with Claude Sonnet 5.5 or GPT-6.2 Sol meanwhile.
  • Retiring and Retired: OpenOrange retires the model on the date shown. Calls to it fail from then.
  • Private, Anonymized and Uncensored: what the provider keeps of a request, and whether the model refuses.
  • Access disabled: you used this model in the period, but may no longer call it.

Limits

  • A model's price is OpenOrange's list price for it. Requests shows what each call actually cost.
  • Calls on your own provider key don't appear on your OpenOrange bill.
  • Routing pools need your Sunport's connection to OpenOrange, and chat models with tools.
  • A model that agents still use can't be deleted. Choose another model for them first.

How-to guides

API keys

An API key lets your own code use models, agents or the OpenOrange API. Each key is a secret shown once, limited to what you chose.

Use it when
Your own code needs models, agents or the OpenOrange API
You get
A secret shown once, limited to what you chose
Cost
The models it calls, at their listed prices
Setup time
One form per key
Needs
Manage everyone's keys for keys on the API page

Four kinds of key

Choose by what your code does:

KeyWhereIt lets your code
Model API keyAPI › Model API keysCall your Sunport's models
Model key of your ownSettings › Personal keys › Model keysCall models under your own limits
Agent API keyAPI › Agent API keysTalk to the agents it names
Personal keySettings › Personal keys › Keys and sign-insUse the OpenOrange API and oo as you

Keys for models

Model keys are OpenAI-compatible. Your code sets the Base URL and the key, then names a model by its alias. The same key also answers Anthropic's Messages API, so tools such as Claude Code work with it.

A model key calls models and nothing else. It may call every model, or only the ones you choose.

  • A model API key is made by an administrator. It can name a User, who can copy the key once, within 24 hours. Copying it needs credentials.manage.
  • A model key of your own is made by you. It works under your own allowed models and budget, as your chat does. Making one needs models.key: administrators have it, and members get it from an administrator.

Keys for agents

An agent API key talks only to the agents it names. With Code access it also uses Code, as its User. A key never does more than the person who made it.

A personal key acts as you in the OpenOrange API and oo. You choose what it can do, which agents, and when it expires.

An agent that runs elsewhere connects with a key of its own. Connected agents explains what it can reach.

Secrets and spend

  • A key's secret is shown once. Copy it before you close the window.
  • Revoke stops a key at once. A revoked key on API can be deleted; its history keeps its name.
  • Each key's requests and spend show on API, and every call on Requests.
  • Administrators see every key's use. Others see their own keys' use.
  • An administrator can give a key a spending budget.

Limits

  • New model keys don't reach realtime voice, batches, files or fine-tuning.
  • Older model API keys made by administrators can call every route of the model proxy. Model key routes on API narrows them.
  • A model key of your own stays yours. Nobody gives it to someone else.

How-to guides

Billing

Billing is how OpenOrange charges your Sunport. You buy credit into a prepaid balance. Calls to models OpenOrange provides, and services, are charged to it.

Use it when
You check the balance, or a charge you didn't expect
You get
A prepaid balance, each call's cost, and records to export
Cost
What your Sunport's model calls and services use
Setup time
None
Needs
Administrator for Billing

The balance

OpenOrange bills your organization from a prepaid balance. Billing shows it as Prepaid balance, or Free trial balance on a free trial.

Each call to a model OpenOrange provides is charged to the balance. So are services such as web search, custom domains and email. Calls on your own provider keys are not: your provider bills those.

When the balance runs low, Billing and Overview warn administrators. By default that's below 20% of your last top-up. When it reaches zero, models OpenOrange provides can pause. Models on your own provider keys keep working.

Topping up

You buy more credit in the OpenOrange app: a card payment in US dollars, with tax added at checkout. Where the app offers it, you can pay every month instead of once. A Sunport with no link to the app tells you to contact OpenOrange instead.

A free trial starts with free credit. When it is used up, Top up for free adds more, while free top-ups are left. After each one, the next waits a while. Buying credit ends the free trial.

Money and Juice Units

The dashboard shows amounts in US dollars. Some oo commands give them in Juice Units (JU): 100 JU is US$1.

Where your Sunport offers it, Settings › Preferences › Currency shows amounts in rubles. That only converts the display: records and exports stay in US dollars.

Where a charge shows

The same charges show at several levels:

  • Requests: every model call, with who made it, the model, its tokens and its Cost.
  • Models: each model's requests and spend in a period.
  • API: each key's requests and spend.
  • Billing: this month's usage by day and by model, and Services this month.
  • Charge records, from Records & export on Billing: one record per charge, exported as a CSV in US dollars.

A call's cost can also read:

  • Own key · no OpenOrange charge: it ran on your own provider key.
  • An amount starting with ~: an estimate from your Sunport's prices, until OpenOrange charges the call.
  • Not charged yet: a recent call OpenOrange hasn't charged yet, with no estimate.
  • Not priced yet: OpenOrange hasn't priced the call. Totals leave it out.

Compute allowance

A paid Sunport includes compute each month: busy vCPU-hours, egress in GB and AWS surplus vCPU-hours. Billing shows this month's use per machine. Going over doesn't slow or block anything, and adds no charge. Turn on Email allowance notifications to email administrators at 80% and 100%.

Limits

  • Billing and Charge records are for administrators. Members see their own requests and spend.
  • An organization OpenOrange doesn't bill sees Estimated spend instead: its Sunport's own estimate, not a charge.
  • A model's listed price is a list price. The charge on Requests is what counts.

How-to guides

Budgets and limits

Budgets and limits cap spending on models and servers. You set them per person, agent, key and model, with one budget for servers.

Use it when
You want spending capped, or a call failed with Budget exceeded
You get
Caps per person, agent, key and model, and a server budget
Cost
Nothing; limits only stop spending
Setup time
One form per limit
Needs
An administrator

Kinds of limit

Each limit holds one thing. A call must pass every limit that applies to it:

LimitHoldsWhere
A person's limitsTheir model calls, from every account and model key of theirsUsers › the person › Limits
An agent's budgetThe agent's model keyAgents › the agent › Models › Budget
A key's budgetOne model API key or agent API keyAPI › the key › Spending budget
Server budgetWhat your Sunport's servers may cost a monthSettings › Billing

The balance holds everything OpenOrange provides. When it reaches zero, those models can pause.

A person's limits

  • Spending caps: a daily, weekly and monthly cap in US dollars. The day and week are the last 24 hours and the last 7 days; the month is the calendar month in UTC.
  • Models: All models, or Only selected. A model can also have caps of its own, inside the overall ones.
  • Access: Blocked stops all their model calls, whatever the caps.

A person who reaches a cap is stopped until spend in that period ages out, or the UTC month rolls over. A model at its own cap stops alone; their other models keep working. A routing pool skips a member at its cap.

An agent's and a key's budget

An agent's budget and a key's budget work alike:

  • A spending limit in US dollars. A limit of 0 blocks new calls.
  • An agent's or a model API key's spend starts again after its reset period, or never.
  • An agent API key counts the last 24 hours, the last 7 days, or the calendar month in UTC.
  • An agent also has requests and tokens per minute, and its allowed models.

An agent API key's budget is shared by every agent on the key. A key with Code write access has no budget.

Server budget

The server budget caps what agent VMs, service VMs, backups and snapshots may cost in a month, at list price. A new server that would pass it is refused until someone raises it. Lowering it never stops servers that run.

Your Sunport's own server isn't counted. VMs on OpenOrange's shared capacity have no list price yet, so they count nothing.

Team limits on the same page cap your team's VMs and published names. It shows only when your team has them. The OpenOrange team approves or declines a request for more.

Limits

  • New calls stop at a limit. Work already running may still add spend.
  • A person's limits take effect within 30 seconds. A server budget change reaches OpenOrange within a minute.
  • A chat account linked to no person gets no person's limits. Link it to a person first.
  • Only administrators set people's, agents' and keys' limits. An agent's key can't change the server budget.

How-to guides

How-to guides

See what a model costs

Before you start: you need See models (models.read). Every member has it.

  1. Open Models.
  2. In Search by alias or provider model, type the model's name.
  3. Open the model.
  4. Read Price and billing.

Check it worked: How it's billed says who charges for the model's calls.

Administrators see every model's use and spend on Models, and with oo models usage.

Models and prices explains each way a model is billed.

Use your own provider key

Before you start: you need Manage everyone's keys (credentials.manage) and Change models (models.write). Not an administrator? Ask one to give you See keys, Manage everyone's keys and Change models.

  1. Open Models and choose Manage providers.
  2. Under Provider keys, choose Add credential.
  3. Choose the Provider and type a Credential name.
  4. Paste the API key, or the fields your provider asks for. Choose Save.
  5. Under Model routes, choose Add model.
  6. Under Credential, choose the key you added.
  7. Find the model in Search catalog, or type its Model ID.
  8. Check the Alias: the name agents and Code use. Choose Save.

Check it worked: in the model's row, choose Test. Health shows OK.

After saving, only a masked hint of the key shows. Rotate replaces it. A key that models still use can't be deleted.

Models and prices explains how your own key is billed.

Create a routing pool

Before you start: you need Change models (models.write). Not an administrator? Ask one to give you Change models.

  1. Open Models and choose Routing pools.
  2. Choose Create routing pool.
  3. Type an Alias: the name people and agents pick.
  4. From Available models, drag a model onto each tier: Simple, Standard, Complex and Reasoning.
  5. Under Fallback model, choose a model.
  6. Choose Save.

Check it worked: the pool shows under Your routing pools. Its alias works once the model proxy refreshes.

To use the pool, pick it under Auto routing in a model picker, or choose it as an agent's model.

Models and prices explains how a pool's calls are charged.

Create a model API key

Before you start: you need Manage everyone's keys (credentials.manage). Not an administrator? Ask one to give you Manage everyone's keys.

  1. Open API.
  2. Under Model API keys, choose Create key.
  3. Type a Name, such as the app that uses it.
  4. Under Models, keep All models, or choose Choose models and pick them.
  5. Choose Create key.
  6. In Copy your new key, choose Copy key, and note the Base URL.
  7. Choose Done.

Check it worked: the key is listed under Model API keys as Active.

Your code calls the Base URL with the key, as an OpenAI client does. Examples shows ready requests for each model.

To make a key for someone else, open Advanced and choose the User. You can let them copy it instead (--hand-off in oo). They copy it once with Show key on its page, within 24 hours. Copying needs credentials.manage.

API keys explains which key to make.

Make your own model key

Before you start: you need Make their own model keys (models.key), signed in as yourself. Not an administrator? Ask one to give you Make their own model keys.

  1. Open Settings and choose Personal keys.
  2. Under Model keys, choose New model key.
  3. Type a Name, after where you use it, such as Laptop.
  4. Choose New model key.
  5. Choose Copy key. The secret is shown once.
  6. Copy the OpenAI-compatible address (base URL) with Copy address.

Check it worked: the key is listed under Model keys.

Without models.key, ask an administrator. They can also make a model API key for you on API.

API keys explains which limits your key follows.

Create an agent API key

Before you start: you need Manage everyone's keys (credentials.manage), and to use each agent the key names. Not an administrator? Ask one to give you Manage everyone's keysand access to each agent.

  1. Open API.
  2. Under Agent API keys, choose Create key.
  3. Type a Name, such as the app that uses it.
  4. Under Agents, choose the agents it may talk to.
  5. Choose Create key.
  6. In Copy your new key, choose Copy key.
  7. Choose Done.

Check it worked: the key is listed under Agent API keys as Active.

Examples shows requests to list agents, send a message and continue a thread, and a reusable skill.md.

To let the key use Code, open Advanced. Set Code access to Read or Run and read, and choose its user. The key then works in Code as that person.

API keys explains what an agent API key can reach.

Create a personal key

Before you start: sign in with your own account. A personal key never does more than you can.

  1. Open Settings and choose Personal keys.
  2. Choose Create personal key.
  3. Type a Name, such as Deploy script.
  4. Under Expires in, choose how long it lasts.
  5. Under What can it do?, choose View only, View and chat with agents or Everything I can do.
  6. Under Which agents?, keep All agents, or choose Choose agents and pick them.
  7. Choose Create personal key.
  8. In Copy your new key, choose Copy key, then I saved the key.

Check it worked: the key is listed under Keys and sign-ins as Active.

For exact permissions instead of these choices, use oo keys create with --permission.

API keys explains when a personal key is the right one.

Revoke a key

Before you start: you need Manage everyone's keys (credentials.manage) for keys on API; an administrator can give you Manage everyone's keys. Your own personal keys and model keys need nothing.

  1. Open API, or Settings › Personal keys for your own keys.
  2. Open the menu at the end of the key's row.
  3. Choose Revoke, then confirm.

Check it worked: the key shows Revoked, or leaves the list. Calls with it are refused.

A revoked key on API can then be deleted with Delete. Its history keeps its name.

API keys explains each kind of key.

Narrow an older model key

Before you start: you need to be an administrator, with Manage everyone's keys (credentials.manage). An administrator can make you one.

  1. Open API and find Model key routes.
  2. Find a key marked Can be narrowed.
  3. Choose Narrow, then confirm with Narrow.

Check it worked: the key shows Model routes only.

A key that made no calls in the last 30 days, or calls other routes, isn't narrowed.

API keys explains what a model key can call.

See what a request cost

Before you start: on Requests, members see their own requests and administrators see everyone's. oo requests needs See requests (requests.read): not an administrator? Ask one to give you See requests.

  1. Open Requests.
  2. Choose the period at the top of the page.
  3. To narrow the list, type in Search by agent or key alias, or choose a kind such as LLM.
  4. Open the request.

Check it worked: Usage shows the request's Cost, Latency and tokens.

On a model or key's page, Open requests lists only that model's or key's requests.

Billing explains each kind of cost a request shows.

Find failed requests

Before you start: on Requests, members see their own requests and administrators see everyone's. oo requests needs See requests (requests.read): not an administrator? Ask one to give you See requests.

  1. Open Requests and choose the period.
  2. Choose Failed.
  3. Open a request to see its details.

Check it worked: each failed request shows its Error, such as Budget exceeded or Rate limited.

Budget exceeded means a spending limit stopped the call.

Budgets and limits explains which limit stops a call.

Check model speed and reliability

Before you start: you need See the Sunport (sunport.read). Not an administrator? Ask one to give you See the Sunport.

  1. Open Observability.
  2. Choose a period of 31 days or less.
  3. Read Failed requests, Median latency and P95 latency.
  4. Under Models, compare each model.

Check it worked: Requests by day charts the calls that succeeded and failed.

Prompt caching shows how much input came from the cache, and what that saved at list prices. Services shows calls to services such as web search; oo observability services needs billing.read.

Billing explains how calls are charged.

Check your balance

Before you start: in the dashboard, you need to be an administrator. With oo, See billing (billing.read) is enough; an administrator can give you See billing.

  1. Open Billing.
  2. Read Prepaid balance, or Free trial balance on a free trial.
  3. Under Usage this month, choose By day or By model.
  4. Read Services this month for services such as web search.

Check it worked: Updated says when OpenOrange last sent the balance.

Who used it links to Overview › Activity, which splits use by person, agent and API key.

Billing explains what the balance pays for.

Top up your balance

Before you start: you need to be an administrator. An administrator can make you one. To pay, you also need to be the Sunport's owner or billing admin in the OpenOrange app.

  1. Open Billing.
  2. Choose Top up in the OpenOrange app.
  3. In the OpenOrange app, choose an amount in whole dollars and pay by card with Stripe.

Check it worked: once the payment is confirmed, it shows under Top-ups and the balance grows.

On a free trial, choose Top up for free once the free credit is used up. With no free top-up left, choose Buy credit in the OpenOrange app; buying ends the trial. A Sunport with no link to the app says to contact OpenOrange instead.

Billing explains what happens when the balance runs out.

Set a low-balance alert

Before you start: you need to be an administrator, with Change billing (billing.write). An administrator can make you one.

  1. Open Billing.
  2. Beside the low-balance alert, choose Edit.
  3. In Warn below, type an amount in dollars, such as 25.
  4. Choose Save.

Check it worked: the alert reads Low‑balance alert: below and your amount.

Turn off under Edit, or oo billing alert --default, goes back to OpenOrange's own warning: below 20% of your last top-up.

Billing explains where the warning shows.

Export your charges

Before you start: in the dashboard, you need to be an administrator. With oo, See billing (billing.read) is enough; an administrator can give you See billing. Your Sunport must be one OpenOrange bills.

  1. Open Billing and choose Records & export.
  2. Choose the period at the top of the page.
  3. Choose Export charges (USD).

Check it worked: your browser downloads a CSV with one row per charge.

Each record links to its request with View request. Unmatched shows charges with no request on this Sunport. One export holds at most 10,000 records; choose a shorter period for more.

Billing explains how charges match requests.

Limit what a person spends

Before you start: you need to be an administrator, with Change billing (billing.write). An administrator can make you one.

  1. Open Users and open the person.
  2. Choose the Limits tab.
  3. Under Spending caps, fill in Daily cap (USD), Weekly cap (USD) or Monthly cap (USD). Leave one blank for no limit.
  4. Under Models, keep All models, or choose Only selected and allow models in the list.
  5. Choose Save.

Check it worked: Saved. Takes effect within 30 seconds. appears.

To stop all of a person's model calls at once, set Access to Blocked, or use --block.

Budgets and limits explains how a person's caps add up.

Limit an API key

Before you start: you need Manage everyone's keys (credentials.manage); an administrator can give you Manage everyone's keys. Only administrators set a key's budget.

  1. Open API and open the key.
  2. For a model API key, under Allowed models, choose All models or Selected models, and pick them.
  3. Under Spending budget, type a Limit (USD) and choose a Budget period.
  4. Choose Save changes.

Check it worked: Changes saved. appears.

An empty limit means no limit. A limit of 0 blocks new calls. An agent API key with Code write access can't have a budget.

Budgets and limits explains how a key's budget works with other limits.

Set the monthly server budget

Before you start: you need Change billing (billing.write), signed in as yourself. Not an administrator? Ask one to give you See billing and Change billing.

  1. Open Settings and choose Billing.
  2. Under Server budget, choose Change.
  3. Type the Monthly budget in US dollars.
  4. Choose Save.

Check it worked: the page says OpenOrange applies the change within a minute, then shows the new budget.

Use the default goes back to your plan's default, as does oo billing server-budget --default.

Budgets and limits explains what the server budget counts.

Ask for higher team limits

Before you start: you need See billing (billing.read) to open Settings › Billing; an administrator can give you See billing. Anyone in your Sunport can ask with oo. Team limits shows only when your team has them.

  1. Open Settings and choose Billing.
  2. Under Team limits, choose Request more.
  3. Type the new totals you need.
  4. In Why your team needs more, say why.
  5. Choose Send request.

Check it worked: Request sent to the OpenOrange team. appears.

One request waits at a time. The OpenOrange team emails the answer to whoever asked.

Budgets and limits explains team limits.