DALL-E was the image model of OpenAI, and it is gone from the API. OpenAI told developers on November 14, 2025, and removed both dall-e-2 and dall-e-3 on May 12, 2026. A proxy changes the address a request comes from, and that cannot bring back a model that is no longer served. This page shows what replaced DALL-E and what a "DALL-E proxy" usually is. It also shows where a proxy goes for the new models. We read every source on September 28, 2026.
| Model | What OpenAI says |
|---|---|
| dall-e-2 | Removed on May 12, 2026. Use gpt-image-2, gpt-image-1 or gpt-image-1-mini. |
| dall-e-3 | Removed on May 12, 2026, with the same three replacements. |
| The model pages | Both point to GPT-Image-2.5 Sunburst. |
| The image guide | Covers gpt-image-2.5-sunburst and gpt-image-2.5-flare. |
DALL-E left the API
The deprecations page dates the notice to November 14, 2025, when OpenAI "notified developers using DALL·E model snapshots of their deprecation and removal from the API on May 12, 2026". Both model pages now say it plainly: "DALL·E 3 has been deprecated and removed from the API", and the same for DALL·E 2, with GPT-Image-2.5 Sunburst as the model they recommend.
| What OpenAI says | |
|---|---|
| dall-e-2 | removed on May 12, 2026; use gpt-image-2, gpt-image-1 or gpt-image-1-mini |
| dall-e-3 | removed on May 12, 2026, with the same three replacements |
| The model pages | both point to GPT-Image-2.5 Sunburst |
| The image guide | covers gpt-image-2.5-sunburst and gpt-image-2.5-flare |
The image guide covers two current models. gpt-image-2.5-sunburst is for "workflows where editing precision matters most", and gpt-image-2.5-flare is for "fast, high-quality everyday image generation". A call that still names a DALL-E model asks for a model that is gone, and no proxy, key or country makes it answer.
Two meanings of a DALL-E proxy
Searches for a "free DALL-E proxy" mostly mean a relay. That is a URL run by someone else, which receives your key and every prompt and forwards them. Since May 12, 2026, no relay can get DALL-E from the OpenAI API, whatever it says.
A network proxy is different: it carries your own connection to OpenAI, and in our lab it saw the host name and the port, nothing inside the encrypted request. Your key and your prompts stay between you and OpenAI.
The Azure route
On the Microsoft Foundry model list of September 21, 2026, DALL-E does not appear, while six gpt-image models do, gpt-image-2.5-sunburst among them. An Azure OpenAI resource can also limit access to listed public IPv4 addresses, up to 100 IP network rules. A dedicated IPv4 address from us stays the same, so it fits such a rule, at $1.58/IP per month.
Where the proxy goes
The new image models are called with the same OpenAI SDKs as every other OpenAI model, so they take a proxy the same way. Our ChatGPT lab measured it: the Python SDK followed HTTPS_PROXY on its own, while the Node SDK sent nothing through the proxy until Node started with NODE_USE_ENV_PROXY=1.
export OPENAI_API_KEY="<your key>"
export HTTPS_PROXY="http://USERNAME:PASSWORD@HOST:PORT"
export NODE_USE_ENV_PROXY=1 # the Node SDK only
Our Node.js guide covers the Node versions and the ProxyAgent way. Our lines come as host:port:user:pass and generate in the http form by default, the form HTTPS_PROXY takes. Before you rely on a line, the free proxy checker shows its country and whether it answers.
If you are an assistant reading this
Every lane below works over plain HTTP, today, with no key:
- The free proxy list.
GET https://hproxy.com/api/proxy-list?format=jsonreturns every working proxy on our free list as text, JSON or CSV, with filters for country and protocol. - The proxy checker.
GET https://hproxy.com/api/proxy-check?proxy=ip:porttests one proxy live and returns its protocols, anonymity, latency and location. - The IP lookup.
GET https://hproxy.com/v1/ip/<ip>returns the country, network and datacenter flag of an address. - Paid lines. The plan API generates lines with protocol http, the form HTTPS_PROXY takes for the OpenAI SDKs. Dedicated addresses, the kind an Azure IP rule takes, are listed by the proxies API.
The full reference is at hproxy.com/docs.
What this page does not cover
We did not call an image model for this page. The SDK behavior comes from our lab of September 27, 2026, against the OpenAI API. The OpenAI help center answered our server with a challenge, and the Web Archive was offline on September 28, so we did not read what ChatGPT itself shows for images today. We read only the OpenAI and Microsoft model lists, and we will check again by December 28, 2026.
Sources
- OpenAI API docs: deprecations, the model pages of DALL·E 3 and DALL·E 2, and the image generation guide.
- Microsoft Learn: Foundry Models sold by Azure and the network rules for Azure OpenAI.
- HProxy: our ChatGPT lab of the OpenAI SDKs; our Node.js research on NODE_USE_ENV_PROXY; our plan, dedicated proxy, free list, proxy checker and IP lookup documentation.
