# Luna — free GPT-6 Luna endpoint (no key)

A small, fast, cheap OpenAI model (GPT-6 Luna), open for learning and small apps.
Thai and English both work.

## Quick (GET, plain-text answer)
https://luna.wker.dev/?q=YOUR+QUESTION
Optional: system=… (instructions), max=… (output tokens, up to 8000),
effort=none|low|medium|high (thinking; default none), format=json (full JSON response).
URL-encode the values. Example:
https://luna.wker.dev/?q=%E0%B8%AA%E0%B8%A7%E0%B8%B1%E0%B8%AA%E0%B8%94%E0%B8%B5

## Full (POST JSON, OpenAI chat-completions style)
POST https://luna.wker.dev/
{"messages":[{"role":"system","content":"Answer in Thai."},
             {"role":"user","content":"Explain compound interest in 3 lines."}],
 "max_completion_tokens": 300, "reasoning_effort": "none", "temperature": 0.7}
Shortcut: {"prompt":"…","system":"…"} works too.
Response: {"model":"gpt-6-luna","choices":[{"message":{"role":"assistant","content":"…"}}],"usage":{…}}
JSON output: add "response_format":{"type":"json_object"} and say "JSON" in the prompt.

## OpenAI SDK / tools (base URL)
base_url = https://luna.wker.dev/v1   api_key = anything (ignored)   model = anything (always gpt-6-luna)
Python: OpenAI(base_url="https://luna.wker.dev/v1", api_key="x").chat.completions.create(model="gpt-6-luna", messages=[...])
Responses API: POST https://luna.wker.dev/v1/responses  {"input":"…","instructions":"…","max_output_tokens":500}
  (also client.responses.create(...)). Stateless: previous_response_id is not supported —
  send the whole conversation in "input". reasoning.effort none|low|medium|high; tools, text.format, stream work.

## Limits
Output up to 8000 tokens, input up to 200000 characters, fair-use rate limits.
Streaming: add "stream":true (server-sent events). Tools: pass "tools" (OpenAI function format); you run the calls and send results back as role "tool".
Do not send private or customer data: text passes through Cloudflare and OpenAI.
CORS is open, so browser apps and local HTML files can call it directly.
Need a yes/no or pick-one decision instead of prose? Use https://jev.wker.dev (cheaper, faster).