Skip to main content
Everything draws down a single unified credit balance (1 credit = 1):theagent,optimizationandbenchmarkingruns,deploys,andper−tokeninference.Coreisacustommonthlyamountyouchoose(1): the agent, optimization and benchmarking runs, deploys, and per-token inference. Core is a custom monthly amount you choose (50-1000)thatconvertstocredits;newaccountsstartwith1000) that converts to credits; new accounts start with 10 free. Active-mode deployments meter their reserved GPU time against the same balance. Every screen shows the same available balance, and top-ups, upgrades, and auto-recharge reflect in it within seconds.
Work that spends GPU time places a temporary hold before it runs, so a run can never overdraw your balance. Agent chat messages hold 1.00eachandsettletotheactualusageofthatturn,typicallyunder1.00 each and settle to the actual usage of that turn, typically under 0.20. Optimization runs hold about twice the estimated charge, with a 5minimum.Whileaholdisactiveyourbalanceshowsa"5 minimum. While a hold is active your balance shows a "X held” note next to it. When the run settles you see exactly what happened: charged amount and released amount. Failed or cancelled runs release the full hold.
The standard LLM unit. A rough rule of thumb: 1 token is about 4 characters of English text. The response returns usage.total_tokens so you never need to guess.
Optimization and benchmarking runs are metered against your unified credit balance. There are no per-plan session quotas. Before a run starts you see the estimated charge and the hold that will be placed (about 2x the estimate, minimum $5). While it runs, the session shows one live “Estimated charge” figure. When it finishes you see the receipt: the charged amount and the released remainder of the hold. You are only charged when the run succeeds.
Credits do not expire while your account is active.
Failed or cancelled runs are refunded to your credit balance in full, automatically.
Settings > Cost shows a daily cost chart, token breakdown, per-model cost, error counts, and your full credit activity. This works on the free tier too: your $10 signup credits and every transaction against them are visible there from day one, and your live balance shows in the credits chip in the top navigation bar on every page. Charts plot the real measured values.
Your prepaid credit balance is the cap: when it runs out, billable requests return 402 until you add credits. To bound recurring spend, set your Core monthly amount at Settings > Billing; for a contractual cap, contact sales about Enterprise.
Credits (1 credit = 1)areyoursingleprepaidbalance,usedforeverything:theagent,optimization,benchmarking,deploys,andinference.Newaccountsstartwith1) are your single prepaid balance, used for everything: the agent, optimization, benchmarking, deploys, and inference. New accounts start with 10 free. Add credits any time at Settings > Cost. Credits do not expire while the account is active.
Volume pricing is an Enterprise conversation, a custom monthly amount above $1000. Contact sales for a quote.
Active mode keeps replicas warm 24/7 for zero cold start. The reserved GPU time is metered against your unified credit balance; exact cost depends on GPU tier and min_replicas, and the Deploy tab shows the projected monthly cost before you commit.
Enterprise customers get monthly invoices with PDF and PO fields at Settings > Billing > Invoices. Contact sales to set this up.
Card on Core (self-serve). ACH, wire, and custom PO on Enterprise.
Yes. Adjust your Core monthly credit amount (50−50-1000) at Settings > Billing; raising it takes effect immediately with prorated billing. To go above $1000/mo, contact sales about Enterprise.

​
Not here?

Plans and pricing

The canonical Core and Enterprise plan comparison.

GPUs and pricing

GPU tiers and per-token rates.

Account and access

Workspaces, members, API keys.

Contact sales

Enterprise quotes and volume deals.