Skip to main content

​
1. What should I use as the API Base URL?

Depending on the protocol, the main options are: Note: If you encounter a 404 error, check whether an extra /v1 was added at the end of the URL. Different tools handle path concatenation differently.

​
2. What should I do if an API call returns 401 “invalid token”?

  1. Confirm that the API Key has been created correctly: https://jiekou.ai/settings/key-management
  2. Confirm that the Authorization format in the request header is: Bearer sk_xxxxxx
  3. If using Claude Code, the environment variable should be set to ANTHROPIC_AUTH_TOKEN=sk_xxxxx (do not add the Bearer prefix; the tool will add it automatically)

​
3. How do I troubleshoot a 404 “page not found” response?

Common causes:
  • Incorrect URL: for example, using /v3/glm-asr but spelling it incorrectly
  • Incorrect model routing: for example, Codex models need to use /v1/responses instead of /v1/chat/completions
  • Automatic path concatenation by tools: some tools (such as cc-switch) automatically add /chat/completions, so the Base URL should not include that path

​
4. How do I configure Jiekou.AI in Claude Code?

Configure the environment variables as follows: Windows cmd: 
set ANTHROPIC_BASE_URL=https://api.highwayapi.ai/anthropic
set ANTHROPIC_AUTH_TOKEN=sk_YOUR_API_KEY
set ANTHROPIC_MODEL=claude-opus-4-1-20250805
set ANTHROPIC_SMALL_FAST_MODEL=claude-sonnet-4-20250514
Mac/Linux bash: 
export ANTHROPIC_BASE_URL=https://api.highwayapi.ai/anthropic
export ANTHROPIC_AUTH_TOKEN=sk_YOUR_API_KEY
export ANTHROPIC_MODEL=claude-opus-4-1-20250805
export ANTHROPIC_SMALL_FAST_MODEL=claude-sonnet-4-20250514
Reference documentation: https://docs.jiekou.ai/docs/integration/claudecode

​
5. How do I resolve Claude Code prompting forced login/verification required?

The latest version of Claude Code may require login. Solutions:
  • Use the Cline extension instead (search for it in the VSCode extension marketplace)
  • Or use it together with forwarding tools such as cc-switch

​
6. How do I call GPT-5.1 / Codex models? Why am I getting a 400 error?

The GPT-5.1 series (including Codex) needs to use OpenAI’s Responses API, not Chat Completions: 
curl "https://api.highwayapi.ai/openai/v1/responses" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer <API Key>" \
-d '{
    "model": "gpt-5.1-codex",
    "input": [...],
    "max_output_tokens": 64000
  }'
The error usually says: Expected ‘thinking’ or ‘redacted_thinking’, but found ‘text’. This is because after the Thinking feature is enabled for Claude 4.5, the context must contain a specific thinking block format. Recommendations:
  • Clear the conversation history and start over
  • Or disable the Thinking feature (if not needed)
  • Use the native Anthropic protocol instead of the OpenAI-compatible protocol (the latter may have incomplete support for Thinking)

​
8. What should I do if the Claude API returns an “input too long” error?

Claude models have input length limits (usually max_tokens supports up to 64000). Please check:
  • Input text length
  • Size of Base64 images (compress oversized images first)
  • Accumulated length of historical conversation context
Contact Support If the FAQ above does not resolve your issue, please contact technical support through the following methods:
  • Enterprise WeChat/WeChat technical support group (recommended, fastest response)
  • Information format to provide:
    • Issue description + screenshot
    • Account ID (UUID)
    • Trace ID (if available, usually included in the error message)
    • Request parameters (after masking sensitive information)