Models / Custom models
Custom AI model API: a base model, a persona and your own model id
A custom AI model API on one prepaid key. Pick a base model from the catalog, give it a persona, default settings and up to 5 fallbacks, and save it under a name you choose. From then on model: "support-bot" is a model id: Cursor, SillyTavern, LangChain, n8n, Claude Code and any OpenAI or Anthropic SDK call it unchanged. No training, no hosting, billed at the base model's price.
Create a custom model, then call it · one PUT, one id
curl -X PUT https://api.uplinkswitch.com/v1/presets/support-bot \
-H "Authorization: Bearer sk-uplink-live-...admin" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-sonnet-5-5",
"models": ["gpt-6.1-sol","gemini-4-argon"],
"params": {"temperature": 0.3, "max_tokens": 800},
"system_prompt": "You are Ada, the support agent for Acme. Answer billing questions in two sentences.",
"system_mode": "prepend",
"description": "Billing support persona"
}'from openai import OpenAI
client = OpenAI(base_url="https://api.uplinkswitch.com/v1", api_key="sk-uplink-live-...")
res = client.chat.completions.create(
model="support-bot", # your custom model id; also @preset/support-bot
messages=[{"role": "user", "content": "Why was I charged twice?"}],
)
print(res.model) # the base or fallback model that answeredimport anthropic
client = anthropic.Anthropic(base_url="https://api.uplinkswitch.com/v1", api_key="sk-uplink-live-...")
msg = client.messages.create(
model="support-bot",
max_tokens=400,
messages=[{"role": "user", "content": "Why was I charged twice?"}],
)
print(msg.content[0].text)# Claude Code, Cursor, SillyTavern, LangChain, n8n: the model field takes the custom id.
export ANTHROPIC_BASE_URL="https://api.uplinkswitch.com/v1"
export ANTHROPIC_AUTH_TOKEN="sk-uplink-live-..."
export ANTHROPIC_MODEL="support-bot"The same id works on /v1/chat/completions, /v1/responses and /v1/messages. The answer's model field and the x-uplink-model header name the model that answered; x-uplink-preset names the custom model. Writes need an admin key or the console; API keys call but cannot change.
What a custom model holds · up to 100 per organization, every change versioned
| Field | What it is |
|---|---|
| Name | 1 to 64 characters: lower-case letters, digits, dots, dashes and underscores. It may not shadow a catalog model or one of your aliases. |
| Base model | Any model id from the live catalog, or one of your org’s aliases. |
| Fallbacks | Up to 5 models tried in order when every route of the one before failed before a byte. |
| Defaults | temperature, top_p, max_tokens, stop and the other request parameters. A value the request sends always wins. |
| System prompt | Up to 32 KB. Prepend puts it ahead of the caller’s system prompt; replace swaps the caller’s out. |
| Tools | Tool declarations the model gets on every call; request tools of the same name replace them. |
| Description | A line for your team and for the model picker. |
What it costs · the base model's price per million tokens, USD, read 2026-11-16
| Base model | Input | Output | Developer list, output |
|---|---|---|---|
| Claude Fable 5.2 | $4.00 | $20.00 | $40.00 |
| Claude Fable 5.1 | $6.00 | $30.00 | $50.00 |
| Claude Opus 5.5 | $2.716 | $9.70 | $20.00 |
| Claude Sonnet 5.5 | $1.20 | $3.00 | $10.00 |
| GPT-6.1 Sol | $0.582 | $2.91 | $10.00 |
| GPT-6 Sol | $0.60 | $3.00 | $10.00 |
| GPT-5.5 | $1.80 | $9.00 | $30.00 |
| Gemini 4 Argon | $0.60 | $3.00 | $10.00 |
A custom model adds nothing to the price: the model that answered bills at its own catalog rate, the base or the fallback that took over. Any of the chat models can be the base, uncensored builds included. The daily real-model check shows which models answered as themselves today.
Every way to name it · the bare name, the @ forms, or a preset field
| "model": "support-bot" | The bare name. Works wherever an app takes a model id. |
| "model": "@preset/support-bot" | The explicit form; the @ prefix on its own also works. |
| "model": "deepseek-v4-pro-0813@preset/support-bot" | The custom model’s prompt, defaults and fallbacks on another base model for this one request. |
| "preset": "support-bot" | A field beside any model, for clients that keep the model field fixed. |
GET /v1/models lists your custom models next to the catalog, so a model picker in Cursor, SillyTavern or n8n shows them by name. The same request shapes work in the OpenAI chat and Responses formats and the Anthropic Messages format.
Create your own AI model API without training · what changes, what stays
Most guides on how to create your own AI model API start with a dataset, a GPU and a serving stack. Most apps that want "their own model" need none of that. They need a model that answers as one persona, follows one set of rules, uses one temperature and one output cap, and keeps answering when a vendor has a bad hour. All of that lives in the request, not in the weights.
A custom model here moves those parts out of your code and onto your key. The system prompt is attached on every call, in prepend mode ahead of whatever the app sends, or in replace mode instead of it. The defaults apply when the request leaves a parameter out; a value the request sets always wins. The fallbacks answer in order when every route of the model before them fails before a byte. Change any of it in the console and every app calling the name gets the new behavior on its next request, with the old version kept in history.
What stays is the base model's knowledge and quality. A custom model built on Claude Fable 5.2 answers like it, at its price, under your name. If the job needs new weights, that is fine-tuning, and this is not it: no training runs here, and nothing you send teaches the model.
Common shapes: a support agent with the company's rules and a strict output cap; a SillyTavern character whose card and settings live on the key, so every front end gets the same persona; a coding model for Cursor or Claude Code with a frontier base and two cheaper fallbacks; an uncensored writing model built on an abliterated base with the house style in the prompt. Each is one PUT and one name.
A custom model keeps its own base model. On an auto router key every other request is routed, and a custom model called by name is not. The routing, fallbacks and presets page covers the request-level forms of the same features, and the SDK guide shows the base URL and key in Python, TypeScript, Go and Rust.
Custom AI model API questions
- One key, every model
- The SDK you already use
- Prepaid, or a monthly plan
- 30-day price lock on every top-up
- Signed receipts for every call