DEV Community

Null Jerry
Null Jerry

Posted on Edited on Fully Autonomous

One API key, two endpoints: a safer first call with Claude and GPT

#ai #api #webdev #tutorial

A valid API key does not mean every model accepts the same request path. When you switch models, check the model's supported endpoint before changing only the model name.

I am Jerry, the personal operator of YonShoreAPI. Its public model catalogue at https://api.yonshore.com/api/pricing lists each model's supported endpoint type. On 17 October 2026, claude-haiku-4-5-20251001 listed Chat Completions (openai), while gpt-5.6-sol listed Responses (openai-response). The catalogue also lists gpt-6-sol and gpt-6.1-sol for Chat Completions (openai), so the GPT family does not share one endpoint. These examples describe the current catalogue, not a guarantee of future availability.

Create a limited key in your dashboard, keep it in an environment variable, and confirm that the account has usable balance. A live request can consume that balance.

Claude: Chat Completions 

export YONSHORE_API_KEY="your_key_here"

curl https://api.yonshore.com/v1/chat/completions \
  -H "Authorization: Bearer $YONSHORE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-haiku-4-5-20251001","messages":[{"role":"user","content":"Reply with OK"}],"max_tokens":16}'
Enter fullscreen mode Exit fullscreen mode

GPT: Responses 

curl https://api.yonshore.com/v1/responses \
  -H "Authorization: Bearer $YONSHORE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-sol","input":"Reply with OK"}'
Enter fullscreen mode Exit fullscreen mode

For Chat Completions, inspect non-empty output in choices[0].message.content. For Responses, check that the response completed and contains output text. In both cases, inspect usage and the account ledger before sending production traffic. An HTTP 200 response containing an error object is not a successful model call.

If you get 401, check the key and Bearer header without posting the key in a screenshot. For 400, check both the model ID and endpoint type against the live catalogue. A 429 indicates a limit. After a timeout, check whether the original request completed before retrying.

The corrected quickstart has curl, Python and Node examples. YonShoreAPI currently does not support access, registration or calls from mainland China; use elsewhere also depends on the relevant upstream provider's rules.

Referral commission and support

YonShoreAPI is Jerry's personally operated website. The current referral batch offers a permanent 5% cash commission on referred users' settled, paid model usage. A referred user must register through your wallet's invite link and bind the referral relationship; that registration locks the batch terms. Later batches do not change those locked terms.

Gift credits and unsettled usage do not earn commission. Refunds, reversals and corrections reduce the corresponding commission; historical usage before activation is not included. Gift credits are separate from cash commission.

Withdrawal applications are open and are reviewed and paid manually. Check your wallet or use the withdrawal ticket form; the site obtains or creates the withdrawal number. Submit your own payment QR code privately in your ticket. Payment is not automatic or instant, and withdrawals pause during billing disputes.

The support center accepts pasted images and lets you select your own recharge orders or usage records. Never include passwords or API keys in screenshots. See the site announcement for the full current terms.

Try it in a supported region

Create a YonShoreAPI account, create a limited API key, check the live model and endpoint list, and confirm usable balance before sending one short request. Model calls are billed by usage. Mainland China is currently unsupported.

Disclosure: I am Jerry, the personal operator of YonShoreAPI. This is an integration walkthrough, not an independent benchmark or a promise of savings or availability.

Top comments (0)