FerryAPI
Low-cost OpenAI-compatible API gateway for DeepSeek, Qwen and Kimi
About FerryAPI
FerryAPI exists because of one useful fact about the model market: the OpenAI API shape has become a de facto standard, and a large number of capable models are substantially cheaper than the frontier ones while being perfectly adequate for the work most applications actually do. Support triage, translation, summarisation and routine automation do not need the most expensive model available. FerryAPI is an OpenAI-compatible gateway to DeepSeek, Qwen, Kimi and MiMo, so switching from an expensive provider to a cheaper one is a base URL change rather than a rewrite.
Several details matter for anyone actually deploying this. Billing runs on a prepaid USD balance with usage billing by token, which avoids both surprise invoices and the currency friction that comes with using Chinese model providers directly. Provider account pools sit behind the gateway, which is what gives resilience when a single upstream account hits a rate limit or goes down. Customer API key management is included, meaning if you are building a product on top, you can issue and control keys for your own users rather than building that layer yourself. Enterprise deployment is available. The target use cases are named directly: support, translation, summaries and automation.
Point your existing OpenAI-compatible client at FerryAPI by changing the base URL and the key, since the request and response shapes are unchanged. Choose among the available models, covering DeepSeek, Qwen, Kimi and MiMo, selecting on cost and capability for each workload rather than routing everything to one model. Top up a prepaid USD balance and usage draws down against it, billed by token, so spend is bounded by what you have deposited. Behind the gateway, provider account pools handle upstream capacity and failover. If you are reselling or building a multi-tenant product, issue and manage API keys for your own customers through the platform rather than implementing key management yourself.
- •OpenAI-Compatible Interface - Existing clients and SDKs work with a base URL change, so migration is not a rewrite
- •Low-Cost Model Access - DeepSeek, Qwen, Kimi and MiMo, substantially cheaper than frontier models for routine work
- •Prepaid USD Balance - Spend is bounded by what you deposit, avoiding surprise invoices and cross-currency friction
- •Token-Based Usage Billing - Costs tracked per token against your balance rather than by opaque request tiers
- •Provider Account Pools - Upstream capacity pooled behind the gateway for resilience against rate limits and outages
- •Customer API Key Management - Issue and control keys for your own end users, so a multi-tenant product does not need its own key layer
- •Multiple Models, One Integration - Route each workload to the model that fits without separate accounts per provider
- •Enterprise Deployment - Dedicated deployment available for organisations that need it
For developers and businesses running high-volume, routine language tasks where frontier-model pricing is not justified: support ticket triage, translation pipelines, summarisation and automation. It suits teams outside China wanting access to Chinese models without navigating provider signup and payment individually. Product builders reselling AI capability get the customer key management layer for free. Anyone currently sending simple classification or summarisation work to an expensive model should compare cost per token here, since that is the entire argument for the product.
Pricing
No pricing page found
Prepaid USD balance with per-token usage billing - no subscription - enterprise deployment available on request
Checked 2026-08-28
