OpenAI · HALO Docs

HALO's OpenAI entrypoint accepts Chat Completions, vision input, GPT Image generations, and GPT Image edits under/openai/v1. The same image routes are also available at the canonical /v1 SDK base.

#モデル

Capability	Production model IDs
Latest chat and vision	gpt-5.6 (Sol alias), gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5
Other chat and vision	gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.2, gpt-5.1, retained GPT-5/GPT-4.1/GPT-4o/o3/o4-mini models
Image generation and editing	gpt-image-2, gpt-image-1.5, gpt-image-1-mini, gpt-image-1, chatgpt-image-latest

#リクエストを作成する

curl https://api.agihalo.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $HALO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5-mini",
    "messages": [{"role": "user", "content": "Hello from HALO"}]
  }'

#OpenAI SDK

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.HALO_API_KEY,
  baseURL: "https://api.agihalo.com/openai/v1",
});

const result = await client.chat.completions.create({
  model: "gpt-5-mini",
  messages: [{ role: "user", content: "Hello from HALO" }],
});

#Vision input

OpenAI image blocks are accepted in user messages as credential-free HTTPS URLs or inline base64 image data URLs. The selected model must declare image-input support.

curl https://api.agihalo.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $HALO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe the important details."},
        {
          "type": "image_url",
          "image_url": {
            "url": "https://example.com/input.png",
            "detail": "high"
          }
        }
      ]
    }]
  }'

Image detail can be auto, low, orhigh. original is limited to the base GPT-5.4, GPT-5.5, and GPT-5.6 models that expose that capability.

#Generate & edit images

curl https://api.agihalo.com/openai/v1/images/generations \
  -H "Authorization: Bearer $HALO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A glass observatory above a moonlit cloud layer",
    "size": "1536x1024",
    "quality": "high",
    "output_format": "png"
  }'

curl https://api.agihalo.com/openai/v1/images/edits \
  -H "Authorization: Bearer $HALO_API_KEY" \
  -F "model=gpt-image-2" \
  -F "prompt=Add a watercolor texture" \
  -F "image[]=@source.png" \
  -F "mask=@mask.png"

Image edits accept one to sixteen JSON HTTPS/data-URL references or multipart image/image[] uploads, plus one optional mask. Generation and edit streams are forwarded without buffering the returned base64 payload.

#制限

OpenAI file_id values are account-bound and unavailable on shared HALO platform routes. Upload the file or use an HTTPS/data URL.
JSON and multipart image request bodies are capped at 64 MB.
Provider-owned files, caches, conversations, tools, unsafe URLs, unknown fields, and unsupported capabilities fail closed.
DeepSeek and custom model-family routes remain text-only.

#ゲートウェイの動作

The gateway validates a priced model, selects an eligible platform key, forwards the canonical request, parses upstream usage, and bills the completed response. GPT Image billing records provider-reported text input, image input, image output, and generated image count. Authentication errors use an OpenAI-style error envelope.

See the current upstream references for models, pricing, and image generation.