CorvusLLM bills against the customer key balance and stops at zero.

Before testing billing

Billing is tied to your CorvusLLM key balance. You only need the dashboard and a small API request to confirm balance changes after a test call.

curl --version

Short answer

CorvusLLM is prepaid. A confirmed payment creates or credits the same customer key, and every delivered API response is deducted from that key balance according to the public model price. When the key reaches zero, requests stop instead of going into debt.

Top-up behavior

Balance rules

• Payment confirmation creates or credits the customer key.
• Top-ups credit the same key, not a second key.
• When balance hits zero, the API returns 429 out of balance.

Pricing formula

Base request cost follows the CorvusLLM prices shown on the public pricing table and model docs.

• Input: requested input tokens x CorvusLLM input price
• Output: completion tokens x CorvusLLM output price
• Cache read: model-specific. Current public factors are 10% of input price for OpenAI/Anthropic routes and 20% for Z.AI routes.
• Anthropic cache write: 125% of input price for 5-minute writes and 200% for 1-hour writes when those buckets are reported.
• Other cache-write billing: only applies when the live catalog exposes a cache-write price for that model.

Cache token behavior

Some provider responses report cache reads or cache writes separately from normal input and output tokens. CorvusLLM prices those buckets from the public model pricing rules shown in the catalog and billing docs.

• Cache read tokens are usually cheaper than normal input tokens, but they still consume prepaid balance.
• Cache write tokens can be more expensive than normal input tokens, especially for long-lived Anthropic cache writes.
• A short user message can still create meaningful cost when the conversation or cached context is large.

Failed request behavior

If a request fails before a model response is delivered, the expected behavior is no customer-facing completion. If a tool reports an error after a provider response was already generated, use the dashboard request history and support channel to inspect the exact case.

Billing safety checklist

1. Confirm the dashboard shows positive balance.
2. Run one small model-list or chat request with the same key.
3. Check that the dashboard request count and balance update after the test.
4. Then run longer prompts, tool calls, or agent workflows.

Related billing resources

Dashboard reflection

The personal dashboard shows the same key, current balance, request count, recent requests, and model breakdown that the runtime writes during real usage.