Billing

What is free

• Basic creation, publishing games and hosting are free. Billing is not on the critical path of free creation and publishing, so a Billing outage does not block publishing.
• Voice input in the site's composers is paid by the platform and never charged to you; each person has a daily allowance of seconds that resets the next day (UTC).
• Running Origin threads on your own computer with Origin Runner costs no sandbox time; model calls are charged as usual.
• POST /v1/messages/count_tokens only counts tokens and is not billed; reading the Meshy animation library is not billed either.

Two balances

Platform spending uses two balances; neither can be cashed out:

• Every paid item belongs to exactly one balance; a subscription plan may include allowances of both, and by default they do not convert into each other.
• When operations publish an Agent-to-Cloud exchange rate, a Cloud Credit shortfall is covered from Agent Credit at that rate, as the last resort.
• Prices, plans and free allowances are published by operations as versions. An item without a published price shows as pending and the API answers feature_pending; it is never treated as free.

How model calls are charged

Calls through the model API, and model calls that Origin, Origin MCP or agents configured by Origin Kit make for you, are charged to the calling account.

• Each model's price is in the model catalog. The unit depends on the model, for example per token, per request or per minute (transcription).
• Before the call the platform holds an estimate; afterwards it settles at the usage the provider reported and returns the rest of the hold.
• With a subscription, the plan's allowance is used first; once it is used up, the wallet is charged only if extra usage is turned on.
• Without enough balance or allowance the request is refused before any model is called, with 403 insufficient_user_quota; a model with no price yet answers 400 model_price_error.
• A generation task whose outcome is unknown (submission_unknown) keeps its hold until the platform reconciles it; it is never retried or refunded automatically.
• /v1/decisions bills input tokens only; the jev-* classifier models bill input tokens and not output.

Origin's cloud sandboxes are charged to Cloud Credit by minutes times class: a running sandbox holds 10 minutes at a time and settles the minutes actually used when it pauses or stops. Without enough Cloud Credit a cloud sandbox does not start; Origin Runner on your own computer is unaffected.

Balance and usage

On the web:

• /me: quota left, recent usage and daily spend.
• /me/usage: four tabs for usage, subscriptions, top-up and free quota; the finance center holds orders, monthly statements, invoices and cost quotes.

Read recent usage with your API token (days from 1 to 90, default 30):

curl "https://api.origingame.ai/billing/v1/me/usage?days=30" \
  -H "Authorization: Bearer $OG_API_KEY"

The answer is in US dollars: currency (USD), window_days, from, to, balance, spent (spend in the window), requests and by_model (model, spent, requests per model, at most 50 models).

Two more balance endpoints accept the API token:

Top-up, subscriptions and redemption

• Top up: the Top up tab of /me/usage. The page shows which payment methods and amounts are available right now.
• Subscriptions: the Subscriptions tab buys plans with your balance and sets your billing preference, extra usage and auto-renewal.
• Redemption codes and referral rewards: the Free quota tab. A redemption code adds quota to your balance directly; no payment is made.
• Points: earned through growth and redeemed at /me/points into Agent Credit or Cloud Credit, with redemption caps that depend on your level. Points cannot be cashed out or converted into player credits.

API token limits

An account API token can carry its own limits, separate from the account balance:

• Spend budget: a token with a budget answers 403 pre_consume_token_quota_failed once the budget is spent; the account balance and other credentials are unaffected.
• Model scope: for a token limited to some models, GET /v1/models lists only those, and other models answer 403.

Manage the token at /me/access; usage is described in Model API.

Bring your own key (BYOK)

You can store your own OpenAI, Anthropic or Google key as a private channel of your account. Text and embedding requests for those models then use your key automatically:

• The model cost is zero; the platform charges only the per-request relay fee operations publish (which may be zero), to Agent Credit. Until a relay fee is published, requests answer feature_pending.
• Only the three official endpoints are used; custom URLs are not supported. Keys are stored encrypted and no endpoint ever returns them.
• When your key fails, the request never falls back to a platform channel or platform funds.

The simplest way is to add, replace or revoke your keys in Origin Kit. You can also call the API directly:

curl https://api.origingame.ai/ai/v1/me/channels \
  -H "Authorization: Bearer $OG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"provider":"anthropic","models":["MODEL_ID"],"secret":"YOUR_PROVIDER_KEY"}'

provider is openai, anthropic or google. GET /ai/v1/me/channels lists channels (metadata only), POST /ai/v1/me/channels/{id}/rotate with {"secret":"..."} replaces the key, and DELETE /ai/v1/me/channels/{id} revokes the channel and erases the ciphertext.