New-API Multi-Model Routing & Billing Strategy

Goal: one OpenAI-compatible entry point backed by multiple authorized upstream channels. Model names, upstream prices and New-API page fields change over time; the configuration below is based on the v1.0.0-rc.49 source code and must be re-verified in the console of the exact version you deploy before going live.

1. Public Model Catalog

Use clear neutral aliases alongside compatibility aliases; never pass off domestic models as OpenAI models.

Public request model	Positioning	Primary channel	Backup channel	Suggested retail price (example)
br-economy / gpt-4o-mini	High concurrency, classification, summarization, general chat	DeepSeek deepseek-chat	Qwen qwen-plus	Input US$0.40/M, output US$1.60/M
br-pt / gpt-4o	Portuguese writing, customer service, localization	Qwen qwen-max	GLM-5	Input US$3.00/M, output US$12.00/M
br-reasoning	High-value complex reasoning, code, long context	GLM-5 or a verified Kimi complex model	DeepSeek deepseek-reasoner	Input US$6.00/M, output US$24.00/M

Compliance red line: gpt-4o-mini and gpt-4o here are only compatibility aliases and do not mean the request is served by OpenAI. The pricing page and terms of service must clearly disclose the real routing strategy; customers should preferably use the br-* names directly.

2. Configuring Upstream Channels

Go to Admin Console -> Channels -> Add Channel and create at least one channel per provider. Obtain API keys only from official platforms, confirm that commercial distribution/resale is permitted, and never use shared keys of unknown origin.

Channel	Type / Base URL	Models field in channel	Model redirect JSON
deepseek-economy	DeepSeek; https://api.deepseek.com/v1	br-economy,gpt-4o-mini,br-reasoning	See below
qwen-ptbr	OpenAI-compatible; DashScope compatible endpoint	br-pt,gpt-4o,br-economy,gpt-4o-mini	See below
glm-premium	OpenAI-compatible; Zhipu official endpoint	br-pt,gpt-4o,br-reasoning	See below
kimi-premium	OpenAI-compatible; Moonshot official endpoint	br-reasoning	Map br-reasoning to the Kimi model actually available on the account

Examples (model names must be adjusted to the values actually available in the upstream console):