One Atlas bill, every modality, already priced

How does Atlas Cloud bill for usage?

Atlas Cloud bills two different ways, and which one you’re on decides what “usage” even means. There’s a Coding Plan, a flat subscription that hands you a usage allowance, and pay-as-you-go, a prepaid balance you draw down request by request. You can hold either or both, and the part that matters for tracking is that the two don’t meter in the same unit: a Coding Plan counts in points, pay-as-you-go counts in money.

	Coding Plan	Pay-as-you-go
Billed as	Flat subscription	Prepaid balance
Metered in	Points	Money
Runs until	A weekly or monthly point cap is hit	The balance is gone
In Atlas’s billing API	No, it’s a fixed fee	Yes, per-model reported cost
In CostCompass	Declare your tier and it folds in	Read on each Refresh

Take the Coding Plan first. It’s a subscription, and the allowance it buys is denominated in points rather than currency. Each request converts to points by a per-model multiplier (roughly your input tokens times an input rate plus your output tokens times an output rate), so a lightweight model spends points slowly while a flagship burns through them at the same request count. Those multipliers vary by model and get revised as models come and go, so treat any you’ve noted as provisional; the Coding Plan page carries the current set. The allowance runs on a dual-limit system (say 120M points across the month alongside 60M in any one week), and Atlas pauses your usage the moment either cap is reached. A Coding Plan is a fixed fee, and the points cap what you can do inside it. They aren’t money that meters out with each call.

Pay-as-you-go is the other option, and the one that will feel familiar from most inference APIs: you fund a prepaid balance and every request draws from it as you go, with no invoice to reason about, just a pool that falls. What makes that pool different is what draws on it. Atlas is a multi-modal gateway, so a text completion, an image generation or edit, and a video render all settle against the same account, each on its own per-model rate: a token model priced by input and output, an image model priced per picture, a video model priced by the second. Three different meters, one balance. That’s the first thing to hold onto on the metered side: an Atlas card is provider spend, not necessarily language-model spend. The number is “what I spent at Atlas this month,” and on a gateway that renders video, one expensive clip can outweigh a month of chat, which is exactly why a single balance is a poor way to watch it, since the level tells you nothing about the mix.

What makes that metered spend easy to track accurately is where the figure comes from. Instead of handing you token counts and a rate sheet, Atlas reports the cost it billed: per model, per day, in money. The pricing has already been done; what you read back is the settled figure, not the ingredients of an estimate. Model names and per-model rates move over time, so treat any figure you’ve memorized as provisional, but the shape holds: Atlas states its own cost.

Which key reads your Atlas Cloud billing?

Before any of that reaches you, there’s one small setup step worth getting right, because a mistyped credential is the most common reason an Atlas connection comes back empty.

When you create a key, it lands in your API Keys list showing a masked value like apikey-d9cb…e968. That masked string isn’t the whole secret: click the reveal (eye) icon to show the full key, or the copy icon to grab it, and paste that into CostCompass. The secret is the credential that authenticates and reads your billing; Atlas sends it verbatim as a bearer token, and it needs read access to your balance and usage. It starts with an apikey- prefix. Unlike consoles that show a key once and then hide it forever, Atlas keeps it revealable from the list, so you can copy it again whenever you need it — which means a connection that comes back empty is almost always a partial copy or the masked placeholder pasted by mistake, not a key you lost your one chance to save.

One connection is enough. Atlas scopes billing to the user behind the key, not to the individual key, so a single Atlas card covers the usage across every key you hold, so you don’t need one card per key.

What does an Atlas Cloud card total, and the Coding Plan it can’t see?

So the metered side of your Atlas bill is straightforward: the per-model cost Atlas reports, summed across every modality, is what you spent on usage. But on Atlas there’s a second charge that the metered side cannot see at all.

The Coding Plan is where the points-versus-money split matters. Because it runs on points against a fixed fee and never draws money from your balance, it appears nowhere in the billing API: not in your balance, and not in the per-model cost feed, both of which are denominated in the money Atlas settles. Left alone, then, the usage number is only half of what you actually pay Atlas: the metered half, with the subscription missing. Closing that gap takes one step. You declare the plan on the Atlas card by its tier name (Coding Starter, Lite, Plus, Max, Ultra, or Enterprise), and CostCompass folds that fee in, so the card totals the whole Atlas bill rather than the part the API will admit to. Tell it which tier you’re on, and the blind spot closes.

This is the same idea a plan-priced provider forces on you (the subscription changes what you owe without ever showing up in the usage meter), reframed here as a gap you close by hand.

How does CostCompass track your Atlas Cloud costs?

Each time you click Refresh, CostCompass asks Atlas for the per-model cost over the current window, takes the reported amount on each model across every modality, and rolls those into your month-to-date total, grouped by model, with the vendor as the header so the breakdown reads cleanly. Because the figures are Atlas’s own settled numbers, CostCompass records them as reported and never reprices them. Today’s day arrives flagged partial and is upserted on each later refresh, so month-to-date stays live and firms up as the day settles. Alongside the usage, it reads your prepaid available balance and folds in any Coding Plan you’ve declared, so the card is the whole Atlas bill from the first refresh on.

Two things make that practical for a solo developer. First, your Atlas secret is encrypted in your browser before anything is stored. The value you paste is sealed with your vault password on your own device, and only the sealed version is ever saved, a blob CostCompass can’t open, because the vault password stays with you. When it’s time to read your billing, your browser unseals the key and passes it to a relay that uses it for that single call to Atlas and is built not to log or keep it. What we hold at rest is locked ciphertext; the key your account runs on never reaches us in a form we can use.

Second, Atlas doesn’t sit alone. The same dashboard rolls its running total up with every other AI and compute provider you’ve connected (Claude, OpenAI, Google, DeepSeek, whatever you route through a fellow one-balance gateway like OpenRouter, a reported-cost provider like xAI, and your hosting) into one figure with a forecast. Reading a stated cost beats inferring spend from a balance drop for the same reason across all of them. Atlas becomes one line in your whole-stack spend rather than a console you check on its own; the ways to track AI costs across providers lay the options out side by side.

How do you forecast next month’s Atlas Cloud bill?

A forecast needs nothing exotic: just a burn rate. Take your spend over the last several days, average it per day, and project that across a full month.

CostCompass does exactly this. It scales your trailing seven-day burn rate to the number of days in next month and adds any subscriptions you’ve declared, including the Coding Plan, producing one forward number for what next month costs if the current pace holds. Seven days is a deliberate window: long enough that a single heavy render doesn’t throw the average, short enough to catch a recent change like a new video workload. Because Atlas reports settled cost, the projection rides on a firm base, and the same engine runs across every provider, so Atlas folds into one whole-stack forecast you don’t add up by hand.

Getting started takes three steps:

1. In the Atlas Cloud console, create an API key, then reveal it in the API Keys list and copy the full secret (the apikey-… value) — the copy icon grabs it for you.
2. Paste it into CostCompass, and if you’re on a Coding Plan, pick your tier so it folds into the card. The secret is encrypted in your browser before it’s stored, so the server only ever holds ciphertext.
3. Click Refresh. CostCompass reads Atlas’s reported per-model cost and your remaining credit, and shows your running total, per-model breakdown, and forecast, with Atlas included in the whole-stack number.