Definition
An AI API gateway is an intermediary service that exposes a single, stable API endpoint in front of multiple large language model providers. Instead of holding separate accounts, keys, balances and SDK configurations for Claude, GPT, DeepSeek, Qwen, Kimi and GLM, a client talks to one base URL. The gateway authenticates the caller, selects the upstream route, forwards the request, and returns the response in the same shape the provider would have returned.
QuantoPort is such a gateway: one key, one balance, and OpenAI-compatible, Anthropic-compatible and Responses endpoints over one host.
What actually happens on one request
- Authenticate. The gateway validates your key and checks the balance attached to it.
- Route. The model name in the request selects the upstream provider and endpoint format.
- Translate. If your request uses the OpenAI schema but the target model speaks the Anthropic schema (or the reverse), the gateway converts the payload in both directions.
- Stream or buffer. Server-sent events pass through unchanged so streaming clients keep working.
- Meter. Input and output tokens are counted from the provider response and deducted from the single balance.
- Report. Usage, per-key spend and remaining credit are visible in the console.
Why teams use a gateway instead of one account per provider
| Concern | One account per provider | One AI API gateway |
|---|---|---|
| Keys | One key per vendor, rotated separately | One key for every model |
| Billing | Separate balances, currencies and invoices | One balance, one top-up |
| Client code | Vendor SDKs and schemas per provider | OpenAI- or Anthropic-compatible base URL |
| Model swap | Code change and new credentials | Change the model name in the request |
| Access | Per-vendor signup, region and payment rules | One onboarding, one payment rail |
How billing works
Billing is per token, priced as a share of the provider’s published rate. At QuantoPort the published shares are:
| Model family | Rate vs official price |
|---|---|
| Claude / GPT | 66% of official rates |
| DeepSeek / Kimi | 88% of official rates |
| GLM-5.3 | 80% of official rates |
| GLM-5.2 | 70% of official rates |
| Qwen | official rates |
Credit packs never expire and the minimum top-up is $10 in USDT over TRC-20, BSC or Solana. The live per-key model list is always available at GET /v1/models, and the base URLs are https://quantoport.com/v1 for OpenAI-compatible clients and https://quantoport.com for Anthropic-compatible clients.
What a gateway does not change
A gateway is not a model. It does not train, fine-tune or host the models it routes to, and it cannot exceed the upstream provider’s context window, rate limits or capabilities. Model names describe compatibility only. QuantoPort is not affiliated with or endorsed by Anthropic, OpenAI, DeepSeek or Alibaba.
Frequently asked
Is an AI API gateway slower than calling the provider directly? A gateway adds one network hop and one payload translation. For streaming chat workloads the first-token latency difference is small relative to model inference time, and the gateway removes the latency you would otherwise spend maintaining several vendor integrations.
Can I move an existing OpenAI or Anthropic client to a gateway? Yes. Point the base URL at the gateway, keep the same request schema, and use a gateway key. No other code change is required for OpenAI-compatible and Anthropic-compatible clients.
How do I know which models are available right now? Query GET /v1/models with your key. The list is per key and is the authoritative source over any static catalogue.
What happens if a provider changes its API? The gateway absorbs the schema change. Your client keeps calling the same endpoint.
Try it with one key
Create a key, point your client at the base URL, and switch models by name.