Inference cost

Cheap inference starts with a bounded workload

Pick the smallest model that reliably completes the task, constrain output length, and measure retries before moving traffic.

Planned recharge methods: Alipay and USDT. Collection is not enabled yet.

Cost levers that matter

  • Model choice
  • Prompt length
  • Maximum output tokens
  • Caching in your application
  • Retry and loop limits
  • Batching where the workload permits

Do not optimize only the token rate

A cheaper model that requires repeated attempts or extensive manual correction can cost more per completed task. Keep a small evaluation set and compare end-to-end outcomes.

Common questions

Where do model availability and prices come from?

Public model pages read the active catalog exposed by the service. A model is not presented as available merely because its name appears in an article.

Can I limit spending by project or API key?

The existing product supports project and key controls, including budget and rate constraints. Exact fields and enforcement behavior are documented in the console and API documentation.

Related resources