Codex Integration for GPT-Image Fails with Incorrect API Key

The Error

Authentication failed
Incorrect API key provided: sk-xxxx****************************A6Af.
You can find your API key at https://platform.openai.com/account/api-keys.
(Request ID: req_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx)

In one sentence: this error is returned by OpenAI’s own servers, not by APIYI. Your code is currently calling api.openai.com, so OpenAI rejects the APIYI key. The key is fine. The request address is wrong.

How to Tell the Request Never Reached APIYI

Either of these two signs settles it:

Sign	Meaning
The error points you to `platform.openai.com/account/api-keys`	This is OpenAI’s standard `invalid_api_key` text. APIYI errors never send you to OpenAI’s website
The request ID is a 32-character string starting with `req_`	OpenAI’s request ID format. You will not find it in APIYI’s logs, because the request never arrived

This is especially common when you ask an AI coding assistant to write the integration. Codex, Cursor, Claude Code and friends see the model name gpt-image-2.5, default to the official OpenAI SDK pattern, and leave base_url at the SDK default https://api.openai.com/v1. The key you paste in is APIYI’s. The two do not match.

Three Fixes, Pick by Situation

① Using Codex / a coding Agent: install the Skills
② Not a coder: hand the prompt to your AI
③ No integration needed: generate on the web

The lowest-effort path is to let the Agent “learn” APIYI before it writes code. Two levels of skill packs are available:

Site-wide skill pack (install this first)

Have your Agent run the command below to install the APIYI skill pack. If that fails, tell it to read https://docs.apiyi.com/skill.md directly:

npx skills add https://docs.apiyi.com

This file is written for AI: base URLs, authentication, model naming rules and common pitfalls. Once installed, the code it writes will point base_url at https://api.apiyi.com/v1 automatically.

Dedicated GPT-Image skill

The GPT-Image-2.5 / 2 Series Agent Skill page ships a ready-to-use Skill: two files and one script covering six models including gpt-image-2.5-flare / gpt-image-2.5-sunburst / gpt-image-2, switched with --model. Text-to-image, multi-image fusion and inpainting are all included.Drop it into Codex, OpenClaw, Claude Code or any coding Agent that can run shell commands, then just say “generate an image of…” You never touch the base URL yourself.

Other image and video models each have an “Agent Skill” page too, filed under that model’s documentation folder. Find the model in the left navigation and look for a sub-page named “Agent Skill”.

If you would rather not learn what a Skill is, copy our ready-made integration prompt to Codex, Claude Code, Cursor or any AI assistant:

Open the GPT-Image-2.5 / 2 Overview
Find the “Let an AI Agent Do the Integration” section and click the copy button on the prompt
Paste it into your AI coding assistant as is

The prompt already hard-codes base_url as https://api.apiyi.com/v1, reads the key from the APIYI_API_KEY environment variable, and pre-empts the usual traps around timeouts, base64 rendering, upload compression and quality parameters. The AI fetches the plain-text version of the docs page first (append .md to any docs URL) and then writes code for your project’s stack.

Every image and video model overview page carries a prompt like this, not only GPT-Image. For the more general three paths (chat Agent, CLI, coding Agent) see the AI Developer Kit.

If you only need images and do not need them inside your own program yet, skip the code entirely:

Copy a key from the “Tokens” page in the APIYI console
Open imagen.apiyi.com and paste the key
Pick gpt-image-2.5-flare (text-to-image) or gpt-image-2.5-sunburst (editing) and generate

The web tool uses the same key and the same API, billed against the same account balance. When you later need it in your app, come back to the first two paths.

Fix It Yourself: One Line

If you already have code generated by Codex, the minimal change is adding base_url to the client. Leave everything else untouched:

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-apiyi-key",
    base_url="https://api.apiyi.com/v1",  # ← add this line
)

result = client.images.generate(
    model="gpt-image-2.5-flare",
    prompt="A shiba inu wearing an astronaut helmet, cyberpunk style",
    size="1024x1024",
    quality="medium",
)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "sk-your-apiyi-key",
  baseURL: "https://api.apiyi.com/v1", // ← add this line
});

const result = await client.images.generate({
  model: "gpt-image-2.5-flare",
  prompt: "A shiba inu wearing an astronaut helmet, cyberpunk style",
  size: "1024x1024",
  quality: "medium",
});

# Override the SDK default without touching code
export OPENAI_API_KEY="sk-your-apiyi-key"
export OPENAI_BASE_URL="https://api.apiyi.com/v1"

Then confirm the request actually reaches APIYI. A model list in the response means you are done:

curl https://api.apiyi.com/v1/models \
  -H "Authorization: Bearer sk-your-apiyi-key"

Follow-up Questions

I changed base_url but still get the same error. Why?

Check in this order:

Multiple config locations: Codex-generated projects often set the URL in .env, a config file and the client constructor. Changing one leaves the others at the default
Environment variable precedence: if OPENAI_BASE_URL is already set to something else on your system, it overrides whatever the code omits. Run echo $OPENAI_BASE_URL to check
No restart: the running process still holds the old config
Spelling: apiyi, not apiyii or apiyl

The simplest proof is the error text itself. As long as platform.openai.com still appears, the request is still going to OpenAI.

Codex says it already switched to APIYI, but the error is unchanged.

Send it the exact error together with this page. Every docs page has a “Copy page” button in the top-right corner. Paste the page content plus the error into the AI and it can pinpoint which config did not take effect. This is the fastest troubleshooting path.

What if the key is still rejected after the request reaches APIYI?

Only then is it time to check the key itself: open the “Tokens” page in the console and confirm the key is enabled, the balance is sufficient and no model whitelist blocks it. Full checklist in Why is my API Key invalid?.

Which model name should I use for gpt-image-2.5?

Default to gpt-image-2.5-flare for text-to-image and gpt-image-2.5-sunburst for editing and inpainting. Both share the same price and parameters. For high-volume, low-cost work use the reverse channel gpt-image-2.5-all. The comparison table on the GPT-Image Series Agent Skill page covers all six.

Why is my API Key invalid?

Why base URL and key must match, with examples in every language.

How to configure Base URL?

/v1 for OpenAI, root domain for Claude, /v1beta for Gemini.

Is there one-click integration?

Hand the docs to your AI coding assistant and let it do the integration.

GPT-Image-2.5 / 2 Overview

Parameters, pricing, integration prompt and common errors.

Basics

Basic API

Image API (Official)

Video API (Official)

Multimodal Understanding API

Text Generation

Audio API (Official)

Retrieval & Safety

Codex Integration for GPT-Image Fails with Incorrect API Key

The Error

How to Tell the Request Never Reached APIYI

Three Fixes, Pick by Situation

Fix It Yourself: One Line

Follow-up Questions

Why is my API Key invalid?

How to configure Base URL?

Is there one-click integration?

GPT-Image-2.5 / 2 Overview

​The Error

​How to Tell the Request Never Reached APIYI

​Three Fixes, Pick by Situation

​Fix It Yourself: One Line

​Follow-up Questions

​Related

Why is my API Key invalid?

How to configure Base URL?

Is there one-click integration?

GPT-Image-2.5 / 2 Overview

The Error

How to Tell the Request Never Reached APIYI

Three Fixes, Pick by Situation

Fix It Yourself: One Line

Follow-up Questions

Related