fujc683/gpt-image-2-6-openai-compatible

选用指南

AI 根据 README 整理 · 2026/11/6 14:17:01

适用场景

适合需要统一接口调用多个图像模型（如GPT Image 2.6）的开发者，可切换模型并自动处理限流回退。

查看 README 依据

[💰 Pricing](https://gola.apimart.ai/k-KPXOMF) · [🔑 Get an API key](https://gola.apimart.ai/k-JAFpQq) --- ## Why a gateway Calling a single model directly means one key per vendor, one failure mode per vendor, and one bill per vendor. A gateway puts them behind one endpoint. | | Direct calls | Through APIMart gateway | |---|---|---| | Keys | one per vendor | **one key** | | Model switch | rewrite code | **change `model` string** | | Failure handling | per-vendor | **unified retry + fallback** | | Billing | scattered | **one balance, one invoice** | | Rate limits | per vendor | pooled |

"cheap": "z-image-turbo", "fallback": "seedream-5-0-flash", } def generate(prompt, tier="default"): try: return call(MODELS[tier], prompt) except RateLimited: return call(MODELS["fallback"], prompt) ``` ## What to watch - **Async by default** — poll `task_id`, do not block on the HTTP call - **Final cost is in `data.cost`**, not the estimate shown at submit time - **Same key across models** — no need to rotate when switching vendors --- ## Related - Prompt library: [gpt-image-2-6-prompts](https://github.com/fujc683/awesome-gpt-image-2-6-prompts)

使用方式

通过OpenAI兼容的API以编程方式使用，需设置base_url和api_key，并调用images.generate接口。

查看 README 依据

## OpenAI-compatible endpoint ```python from openai import OpenAI client = OpenAI(base_url="https://api.apimart.ai/v1", api_key="YOUR_KEY") img = client.images.generate( model="gpt-image-2.6-flare", prompt="editorial studio portrait, hard key light, grey backdrop", n=1, ) print(img.data[0].url) ``` The request shape is identical to OpenAI's `images.generate`, so existing code keeps working — only `base_url` and `model` change. ## Multi-model routing ```python MODELS = { "default": "gpt-image-2.6-flare", "quality": "gpt-image-2.6-sunburst",

部署要求

需要一个API密钥（来自APIMart）以及可访问该服务的环境。

查看 README 依据

[💰 Pricing](https://gola.apimart.ai/k-KPXOMF) · [🔑 Get an API key](https://gola.apimart.ai/k-JAFpQq) --- ## Why a gateway Calling a single model directly means one key per vendor, one failure mode per vendor, and one bill per vendor. A gateway puts them behind one endpoint. | | Direct calls | Through APIMart gateway | |---|---|---| | Keys | one per vendor | **one key** | | Model switch | rewrite code | **change `model` string** | | Failure handling | per-vendor | **unified retry + fallback** | | Billing | scattered | **one balance, one invoice** | | Rate limits | per vendor | pooled |

## OpenAI-compatible endpoint ```python from openai import OpenAI client = OpenAI(base_url="https://api.apimart.ai/v1", api_key="YOUR_KEY") img = client.images.generate( model="gpt-image-2.6-flare", prompt="editorial studio portrait, hard key light, grey backdrop", n=1, ) print(img.data[0].url) ``` The request shape is identical to OpenAI's `images.generate`, so existing code keeps working — only `base_url` and `model` change. ## Multi-model routing ```python MODELS = { "default": "gpt-image-2.6-flare", "quality": "gpt-image-2.6-sunburst",

已知限制

API调用是异步的，需轮询任务ID，且最终费用以data.cost为准。

查看 README 依据

"cheap": "z-image-turbo", "fallback": "seedream-5-0-flash", } def generate(prompt, tier="default"): try: return call(MODELS[tier], prompt) except RateLimited: return call(MODELS["fallback"], prompt) ``` ## What to watch - **Async by default** — poll `task_id`, do not block on the HTTP call - **Final cost is in `data.cost`**, not the estimate shown at submit time - **Same key across models** — no need to rotate when switching vendors --- ## Related - Prompt library: [gpt-image-2-6-prompts](https://github.com/fujc683/awesome-gpt-image-2-6-prompts)