Billing

Billing is prepaid. You top up your org's balance, and OpenResearch deducts compute, storage, and cloud-agent charges as you use them. Charges are passed through at cost — you pay the underlying provider and infrastructure rates, with no OpenResearch markup. A non-positive balance blocks new hosted compute and tears down your running hosted instances until you top up.

Where to find it

Open Org settings > Billing. The page shows your current balance, in-flight charges, current storage usage, and recent activity. Click Top up to add funds via hosted checkout.

What you're charged for

• Hosted compute — billed per millisecond a sandbox is running. Charges accrue continuously and post in whole-cent slices.
• Storage — logs and artifacts kept in R2. The first 50 GB per organization is free; anything above that is billed at $0.015/GB-month on the overage only. Measured against your current footprint, accrued continuously.
• Cloud agent — the model tokens the autoresearch assistant spends on your behalf, billed at the model provider's upstream prices. See How cloud-agent billing works below for the full breakdown.

How cloud-agent billing works

When you chat with the in-project assistant — or it drives experiments autonomously — it calls a large language model on your behalf. You're charged for the input and output tokens of those model calls, priced at the model provider's published per-token rates (e.g. Anthropic for Claude models, OpenRouter for others) and passed through to you at cost.

• Input tokens are everything the model reads each turn: your message, the project context, the agent's instructions, and the conversation history so far. Output tokens are what the model generates back, including the reasoning and tool calls it makes while working.
• The price per token depends on the model in use — more capable models cost more per token. The autoresearch assistant runs many steps per turn, and the conversation history it re-reads grows over a session, so a long autonomous run costs more than a single short question.
• Charges post per turn, after the turn completes, and appear in the activity log grouped by project and day. The amount is the provider's token cost for that turn, converted to your balance currency. Expand a day's entry to see each turn's model and its input/output token counts. (Entries from before this breakdown shipped show the daily total only.)

This is separate from hosted compute: when the agent launches an experiment on a GPU or CPU instance, that instance's running time is billed as compute (above), while the agent's own model calls are billed as cloud-agent tokens. A turn can incur both.

Self-managed instances (RunPod, Lambda, anything you SSH into) are billed by your provider directly, not by OpenResearch — only the logs and artifacts they produce count toward storage.

The In flight tile on the billing page shows an estimate of charges accumulating on currently-running sandboxes since their last billed tick — useful when you want to know what a long run is costing you in real time.

Activity log

The Recent activity section shows every billing event: top-ups, compute and storage charges (with the time window they cover), cloud-agent charges, refunds, and manual adjustments. Each entry shows the amount; positive entries add to your balance, negative ones deduct.

Running out of balance

OpenResearch checks your balance when you try to provision new hosted compute — running an experiment or cold-starting the cloud agent. A non-positive balance blocks the request and surfaces a top-up link.

It also reaps what's already running. A billing tick runs every few minutes; once it posts enough charges to push your balance to zero or below, it tears down every hosted instance it bills you for. The charge that tips you over is caught on that same tick, so a long run can leave you slightly negative before its box is shut down — top up before you cut it close. Self-managed instances (RunPod, Lambda, anything you run on your own provider account) aren't billed by balance and are never reaped — your provider keeps charging you until you stop them yourself.