API Reference

Public, machine-readable model list and prices. No API key is required. One entry per public model with model_group, mode, input_cost_per_token and output_cost_per_token (USD per token: multiply by 1,000,000 for the per-1M price), context limits, shared rpm / tpm capacity and capability flags. A model served from several pools shows its highest pool rate (a Data Zone rate is up to 10% above Global); the exact charge for a call is always its x-litellm-response-cost header. The old GET /v1/pricing endpoint is retired.

curl https://ggpt.aiforbiz.email/public/model_hub
# [{"model_group":"gpt-5.6-luna","mode":"chat","input_cost_per_token":2.2e-07,
#   "output_cost_per_token":1.32e-06,"max_input_tokens":922000,"rpm":12064,"tpm":12064000, ...}, ...]

List the models available to your key.

curl https://ggpt.aiforbiz.email/v1/models \
  -H "Authorization: Bearer $AIFORBIZ_KEY"

Chat and reasoning. Models: gpt-6.1-sol, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-chat-latest, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro, gpt-6.1-terra. (gpt-5.4-pro is a Responses-API model: call it via /v1/responses below; the OpenAI-style /v1/chat/completions is auto-translated, but an Azure-SDK .chat.completions call returns 404.)

curl https://ggpt.aiforbiz.email/v1/chat/completions \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [
      {"role": "user", "content": "Summarize relativity in two sentences."}
    ],
    "max_completion_tokens": 500
  }'

Responses API. Required for gpt-5.4-pro (Azure does not expose chat completions for this model). Use your SDK's client.responses.create(...) or POST directly.

Built-in image_generation tool. gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna can generate and edit images inline, using the same request shape as OpenAI. Set the tool's model to gpt-image-2.5-flare or gpt-image-2.5-sunburst; action, quality, size, background, output_format, moderation and max_tool_calls all apply. The image arrives as base64 in an image_generation_call output item, and the image tokens are reported separately under tool_usage.image_gen and billed at the GPT Image 2.5 rates above, on top of the orchestrating model's own tokens.
Required: use https://ggpt.aiforbiz.email/v1 for image-generating requests and set your client timeout to ≥ 300s. stream: true is not supported together with this tool.

curl https://ggpt.aiforbiz.email/v1/responses \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-luna",
    "tools": [{
      "type": "image_generation",
      "model": "gpt-image-2.5-sunburst",
      "quality": "low",
      "action": "edit",
      "moderation": "low"
    }],
    "max_tool_calls": 1,
    "input": [{
      "role": "user",
      "content": [
        {"type": "input_text", "text": "Mount the picture flat on the wall."},
        {"type": "input_image", "detail": "low", "image_url": "https://your-host/photo.png"}
      ]
    }]
  }'

curl https://ggpt.aiforbiz.email/v1/responses \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-astra",
    "input": "Write a Python function that reverses a linked list."
  }'

Generate an image. Models: gpt-image-2.5-sunburst, gpt-image-2.5-flare, gpt-image-2. The GPT Image 2.5 models are limited to 19 requests per minute each and need explicit authorization. Returns base64 image data.
Required: use https://ggpt.aiforbiz.email/v1 for every image-generation request and set your client timeout to ≥ 300s.

curl https://ggpt.aiforbiz.email/v1/images/generations \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A watercolor fox in a misty forest",
    "size": "1024x1024",
    "n": 1
  }'

Edit a PNG or JPEG hosted at a public HTTPS URL with gpt-image-2, gpt-image-2.5-flare, or gpt-image-2.5-sunburst. Send JSON with exactly one of image_url or image_urls. The gateway authenticates the key before downloading, blocks private/internal destinations, validates the image, and then sends its bytes through the normal LiteLLM/Azure edit path.
Required: use https://ggpt.aiforbiz.email/v1 and set the client timeout to ≥ 300s.

curl https://ggpt.aiforbiz.email/v1/images/edits \
  -H "Authorization: Bearer ***" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "image_url": "https://cdn.example.com/input.png",
    "prompt": "Change the background to a beach",
    "quality": "medium"
  }'

curl --max-time 300 https://ggpt.aiforbiz.email/v1/images/edits \
  -H "Authorization: Bearer $LITELLM_API_KEY" \
  -F model=gpt-image-2.5-flare -F image=@input.png \
  -F prompt="Change the background to a beach" \
  -F quality=low -F size=1024x1024

Use gpt-image-2.5-sunburst for Sunburst. Omit response_format. Existing multipart uploads remain supported on the same endpoint. URL inputs must use HTTPS on port 443, resolve only to public addresses, return a valid PNG/JPEG, and stay within the 20 MiB combined input limit. Optional URL fields: image_urls (up to four images) and mask_url. These models return data[].b64_json; Azure does not provide URL output. The JSON adapter returns one edited image per request.

Adapter errors: 400 unsafe/unreachable URL, 401/403 auth/access, 408 download timeout, 413 size/pixels, 415 invalid image, and 429 budget/rate limit, or a queue wait that would pass the limit (see Image queue and batches). Azure edit and safety errors pass through.

Image queue and batches. Image calls never fail just because a model is at its Azure rate limit. /v1/images/generations and /v1/images/edits (JSON and multipart) wait in a fair line per model, one turn per API key, and go out as soon as Azure has room. The response is the normal image response plus the header x-aiforbiz-queue-wait-ms (time spent waiting). A call waits up to 60 s on ggpt.aiforbiz.email and up to 240 s on ggpt.aiforbiz.email; if the wait would pass that, you get 429 with a Retry-After header. If you have several images, or a call that may wait, submit a batch instead and collect the results later. Closing the connection removes a waiting call from the line.

Submit 1 to 10 items for one model. Each item is a generation (prompt, optional size, quality, n, background) or an edit (the same fields plus image_url or image_urls, optional mask_url, public HTTPS URLs as above). Access and the model are checked once, up front. The reply is 202 with the batch id (imgb_...) and an item list. Every item is still billed to your key like a normal call.

curl https://ggpt.aiforbiz.email/v1/images/batches \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "items": [
      {"type": "generation", "prompt": "A watercolor fox in a misty forest", "size": "1024x1024", "quality": "low"},
      {"type": "edit", "prompt": "Change the background to a beach", "image_url": "https://cdn.example.com/input.png"}
    ]
  }'

# → 202 {"id":"imgb_...","object":"image.batch","status":"in_progress","model":"gpt-image-2",
#            "created_at":1760000000,"expires_at":1760003600,"items":[{"index":0,"status":"queued"}, ...]}

Poll a batch. The batch status is in_progress while any item is queued or running, then completed (at least one image succeeded), failed or cancelled. Each item has its own status (queued, running, succeeded, failed, cancelled). A succeeded item carries result (the usual image response with data[].b64_json) and cost in USD (the same figure as x-litellm-response-cost); a failed item carries error.message. Add ?results=false to poll statuses without the image data. Poll every few seconds.

curl https://ggpt.aiforbiz.email/v1/images/batches/imgb_... \
  -H "Authorization: Bearer $AIFORBIZ_KEY"

Cancel the items that have not started. Items already running finish and keep their result.

Vector embeddings. Models: text-embedding-3-large, text-embedding-3-small.

curl https://ggpt.aiforbiz.email/v1/embeddings \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "text-embedding-3-large",
    "input": "the quick brown fox"
  }'

Speech-to-text. Model: gpt-transcribe. Multipart upload of an audio file; returns {"text": ...}.

curl https://ggpt.aiforbiz.email/v1/audio/transcriptions \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -F "file=@meeting.wav" \
  -F "model=gpt-transcribe"

Decisions API. Ask typed questions about text or images and get back probabilities, not prose: predicate (yes or no), choice (pick one of the listed values) and score (a level scale). Use it for classification, routing and moderation. Model: gpt-6-luna only; your key must be allowed that model. input is a string or a message array with input_text and input_image parts. Limits per request: up to 400 questions, 128 images and 32 MB. The reply is Azure's, unchanged: answers (one per question, in order) and usage.

curl https://ggpt.aiforbiz.email/v1/decisions \
  -H "Authorization: Bearer $AIFORBIZ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-luna",
    "input": "The delivery was late and the box was damaged.",
    "questions": [
      {"type": "predicate", "name": "negative",
       "instructions": "Is the writer unhappy?"},
      {"type": "choice", "name": "topic",
       "instructions": "What is the main complaint?",
       "choices": [{"value": "delivery"}, {"value": "damage"}, {"value": "price"}]},
      {"type": "score", "name": "anger",
       "instructions": "How angry is the writer?",
       "levels": [{"label": "low"}, {"label": "high"}]}
    ]
  }'

Each answer carries its probabilities, for example {"type": "predicate", "name": "negative", "probability": 0.99}, or for a choice {"type": "choice", "name": "topic", "choice": "delivery", "probabilities": [...], "confidence": 0.97}.

Price. Only input tokens are billed; there is no output charge (usage.output_tokens is always 0). Global rate: $0.10 per 1M input tokens, $0.01 per 1M cached input (a small request of about 170 tokens costs about $0.00002). A call answered by a Data Zone account is billed at that zone's gpt-6-luna rate (US $0.11, EU $0.12 per 1M input). Above 272k input tokens the long-context rates apply. The real usage is charged to your key like any other call and the cost is in the x-litellm-response-cost response header. Your budget and your key's gpt-6-luna request and token limits apply (429 with Retry-After when reached; 400 when the budget is spent). Calls move between accounts automatically when one is busy; a 429 or 503 means every account was busy, so retry shortly.

Your key's budget and running spend, so you can show remaining balance in-app without opening the dashboard. Call it with just your key and no parameters; it returns only your own key. Note this route is at the root, not under /v1.
Budget left = max_budget − spend (USD). A null max_budget means the key is uncapped.

curl https://ggpt.aiforbiz.email/key/info \
  -H "Authorization: Bearer $AIFORBIZ_KEY"

# → {"key":"...","info":{"spend":0.42,"max_budget":100.0,"soft_budget":null, ...}}
# remaining budget = max_budget - spend = 99.58 USD

# Python: remaining budget in a couple of lines
import requests
info = requests.get("https://ggpt.aiforbiz.email/key/info",
                    headers={"Authorization": f"Bearer {AIFORBIZ_KEY}"}).json()["info"]
remaining = None if info["max_budget"] is None else info["max_budget"] - info["spend"]
print(remaining, "USD left")

Realtime voice and text over WebSocket (OpenAI Realtime API, GA protocol) for the gpt-realtime-2* models. Connect with your key in the Authorization: Bearer header and the model in the query string; the server sends session.created, you configure the session with session.update, add turns with conversation.item.create and ask for an answer with response.create. For voice, set "output_modalities": ["audio"], stream microphone audio with input_audio_buffer.append (base64, 16-bit PCM, 24 kHz mono) and play the response.output_audio.delta chunks. Use wss://ggpt.aiforbiz.email for calls longer than about 100 seconds. Billed per text and audio token at the prices above.

# pip install websockets
import asyncio, json, os, websockets

URL = "wss://ggpt.aiforbiz.email/v1/realtime?model=gpt-realtime-2.1"
KEY = os.environ["AIFORBIZ_KEY"]  # your gateway key

async def main():
    async with websockets.connect(URL, additional_headers={"Authorization": f"Bearer {KEY}"}) as ws:
        print(json.loads(await ws.recv())["type"])          # session.created
        await ws.send(json.dumps({"type": "session.update", "session": {
            "type": "realtime", "output_modalities": ["text"],
            "instructions": "You are a helpful call-center assistant. Be brief."}}))
        await ws.send(json.dumps({"type": "conversation.item.create", "item": {
            "type": "message", "role": "user",
            "content": [{"type": "input_text", "text": "Say hello in one sentence."}]}}))
        await ws.send(json.dumps({"type": "response.create"}))
        async for message in ws:
            event = json.loads(message)
            if event["type"] == "response.output_text.delta":
                print(event["delta"], end="", flush=True)
            elif event["type"] == "response.done":
                print(); break
            elif event["type"] == "error":
                print(event["error"]); break

asyncio.run(main())