AI API Gateway

Call models from several providers with one key that starts with sk-nc-. Use the OpenAI, Anthropic or Gemini API format, and pay from your account balance for the tokens you actually use.

How to connect

Point your SDK's base URL at this site and use an sk-nc- key created in the console. Any of the three formats can call any model in the table; when the formats differ, the gateway converts the request and the response.

• OpenAI format: POST /v1/chat/completions with the key in the Authorization: Bearer header. /v1/embeddings, /v1/images/generations and /v1/models are also available.
• Anthropic format: POST /v1/messages with the key in the x-api-key header.
• Gemini format: POST /v1beta/models/{model}:generateContent (or :streamGenerateContent for streaming) with the key in the x-goog-api-key header or the key parameter.

Features

• Formats work across models: OpenAI, Anthropic and Gemini SDKs can all call any model in the table. Matching formats are forwarded as they are; otherwise the gateway converts the request and the response.
• Streaming: All three formats support streamed responses. The cost is calculated from actual token usage when the request finishes.
• Embeddings and images: /v1/embeddings returns vectors and /v1/images/generations creates images. Models priced per image or per second are marked in the table.
• Automatic failover: When a model is offered through several connections and one cannot be reached, returns a 5xx error or is rate-limited, the gateway moves on to the next one, so your client does not have to retry.
• Monthly limit per key: Each key can have its own monthly spending limit. Once a key reaches it, that key's requests are refused while other keys keep working.
• Per-request logs: Every request records the model, token usage and cost. View them by key and model under Usage & Logs.
• Cache and long context priced separately: Cache reads and cache writes have their own prices. For some models, a prompt over a threshold is billed entirely at the long-context price; the table shows the threshold and the price.
• Refused when the balance is short: Before a request is sent, an amount covering the model's maximum output is reserved. If your balance cannot cover it, the request is refused up front instead of running into debt.

Get started

1. Sign up and top up: Create an account and add funds to your balance in the console.
2. Create an API key: Create an sk-nc- key under AI Services › API Keys. You can set a monthly limit.
3. Change the base URL: Point your SDK's base URL at this site, enter the key, and use the model names from the price table.
4. Check your usage: Usage and cost for each request are under Usage & Logs; charges are recorded in Finance.

Billing rules

• Prices are per 1M tokens, with input, output, cache read and cache write priced separately. Models billed per image, per second or per call are marked separately.
• When a model is offered through several connections at different prices, each request is billed at the connection that served it. The table shows the lowest price and marks it "and up".
• For models with an off-peak price, whether a request falls in peak hours depends on when it starts. Web search is charged per call and gets no off-peak discount.
• A request whose prompt exceeds the long-context threshold is billed entirely at the long-context price.
• Before a request is sent, an amount covering the model's maximum output is reserved. If your balance or the key's monthly limit cannot cover it, the request is refused.
• Usage is deducted from your account balance about once a minute and summarized as one bill in Finance every hour.
• Requests the upstream rejects with an error are not charged.
• Parameters such as service_tier and speed modes are ignored; every request is processed and billed at the standard price.

Model prices

Models available: 6 · Model providers: 2

Model	Context	Input	Output	Cache read
DeepSeek-V4.1-Flash deepseek-flash	1M	$0.21 and up (Official $0.30)	$0.84 and up (Official $1.20)	$0.0042 and up (Official $0.006)
DeepSeek-V4-Pro deepseek-v4-pro	1M	$0.924 and up (Official $1.32)	$2.772 and up (Official $3.96)	$0.0308 and up (Official $0.044)
GPT-6 Astra gpt-6-astra	1M	$10.30 (Official $10.00)	$51.50 (Official $50.00)	$1.03 (Official $1.00)
GPT-6.1 Sol gpt-6.1-sol	1M	$2.06 (Official $2.00)	$10.30 (Official $10.00)	$0.103 (Official $0.10)
GPT-6.1 Terra gpt-6.1-terra	1M	$1.03 (Official $1.00)	$5.15 (Official $5.00)	$0.0515 (Official $0.05)
GPT-6 Luna gpt-6-luna	1M	$0.103 (Official $0.10)	$0.515 (Official $0.50)	$0.0103 (Official $0.01)

Prices are in US dollars. When a model is offered through several connections at different prices it is marked "and up", and each request is billed at the connection that served it. The Model Hub in the console and your request logs are authoritative.

Frequently Asked Questions

Which models can I call?

The price table lists every model you can call right now and updates automatically as connections and model settings change. The Model Hub in the console shows the same list.

How do the prices compare with the official ones?

The provider's official price is shown under each price in the table. Our price is the official price multiplied by the connection's pricing ratio, so you can compare models and connections directly in the table.

I use the OpenAI SDK. What do I change?

Only the base URL and the key: set base_url to this site's address plus /v1, set api_key to your sk-nc- key, and use the model names from the price table.

Can I call Claude or Gemini with the OpenAI format?

Yes. Any of the three formats can call any model in the table. When the formats differ, the gateway converts the request and the response; when they match, it forwards them as they are.

My balance looks sufficient, so why was the request refused?

Before a request is sent, an amount covering the model's maximum output is reserved, and your balance has to cover it. Also check whether the key has a monthly limit and whether it has been used up.

Am I charged for failed requests?

Requests the upstream rejects with an error are not charged. The request log shows the status and cost of every request.

Do you support service tiers such as Batch or Priority?

No. service_tier and speed-mode parameters are ignored, and every request is processed and billed at the standard price.

How do I control spending?

Give each key a monthly limit; once it is reached, that key's requests are refused. Use a separate key for each project so you can see their usage separately.

Related pages: Subscriptions · KYC Service · Proxy Subscription · AI API Gateway