As recorded at collection: Sub2API AI model APIs available for free
As recorded at collection: Shangan9 (上岸9) Sub2API is a third-party relay API service that provides official APIs on their behalf, and offers ChatGPT, Claude, Gemini, and Codex family models for free. You need an API key issued directly by the site.
- What you get
- Free API access to GPT·Claude·Gemini·Codex family models
- Price
- Free
- Valid until
- To be confirmed
- Requirements
- API key issued directly by the site required · Do not run load tests with a large number of concurrent requests from the start · Do not make endless requests for long periods · If a particular model is unstable, switch to another model or try again later
Details
Shangan9 (上岸9) Sub2API is a relay API service that lets you use AI models for free by connecting them to programs that support the OpenAI format. It can be used with chat apps such as Cherry Studio, NextChat, and LobeChat, and development tools such as Cursor, Claude Code, and Codex CLI. The service says it is also suitable for large-scale testing, automation scripts, and light production use.
After getting an API key (an authentication string required to use the service) from the site, enter https://api.shangan9.cc.cd/v1 in the address field (Base URL) of the program you use and paste the issued key into the key field. For programs that require the full address, enter https://api.shangan9.cc.cd/v1/chat/completions. Select a model name from the site's model list. You can connect existing AI apps and development tools using these settings.
The operator says it keeps operating costs very low and asks users to avoid excessive use, not to run load tests with a large number of concurrent requests from the start, and not to make endless requests for long periods. It advises checking the connection with a lightweight model before trying more advanced models one by one, and checking small tasks in development tools before running large projects. A particular model may be unstable; if so, switch to another model or try again later. Users are also asked to provide the model name, time of occurrence, and error details when reporting an error.
Sources and verification history
- Collection channel
- NodeLoc
- Collected at
- July 3, 2026, 16:12 KST
- First published
- Updated
- Source
- nodeloc.com/t/topic/97465
- Updates
- July 26, 23:23 · Reclassified due to criteria change