Deep Research · July 13, 2026

Can the ChatGPT/Codex Subscription Generate Images in ForgeBot’s Context?

Yes. The earlier “Codex OAuth is text-only” conclusion is outdated. Current Codex exposes native image generation to ChatGPT-authenticated sessions, and ForgeBot can invoke that capability from Hermes through the Codex CLI.

YES

Use OpenAI GPT Image 2 through Codex’s native image_gen tool while logged in with ChatGPT.
This uses the ChatGPT/Codex subscription path rather than OpenRouter pay-as-you-go credits or an OpenAI Platform API key.

External Evidence

OpenAI’s Codex issue requesting subscription-backed image generation was closed as completed on April 16, 2026. An OpenAI maintainer stated: “This is now supported if you're logged in with a ChatGPT account.”
Current Codex source marks image_generation as stable and enabled by default.
Current Codex source pins the native renderer to gpt-image-2.
The runtime gate requires OpenAI actor authorization or Codex-backend authentication plus an image-capable provider/model.

Internal Evidence

Installed Codex: 0.142.4.
Authentication: Logged in using ChatGPT.
Feature state: image_generation stable true.
A controlled run explicitly removed OPENAI_API_KEY and OPENROUTER_API_KEY, then successfully generated a real PNG through native image_gen.
Verified artifact: 1254 × 1254, RGB PNG, 1,625,703 bytes.

What This Means—and What It Does Not

Question	Answer	Implication
Can the subscription generate images?	Yes, through Codex native image generation.	ForgeBot can call Codex CLI/app-server from Hermes.
Is this a general OpenAI Images API credential?	No.	Codex OAuth/access tokens are scoped to Codex workflows. General API calls still require a separately billed Platform API key.
Which image model is used?	`gpt-image-2` in current Codex source.	The orchestration model may be GPT-5.5/5.6, but the image renderer is GPT Image 2.
Is it unlimited?	No guarantee.	Usage is subject to ChatGPT/Codex plan limits and fair-use controls, but it does not consume OpenRouter credits.
Can Hermes’ existing `openai-codex` provider call it directly?	Not through Hermes’ current model-provider interface.	Use Codex CLI/app-server as the bridge, or add a dedicated integration later.

Recommended Routing

Primary: Codex native image_gen with ChatGPT login.
Secondary: Nous-managed GPT Image 2 if Codex is unavailable.
Paid fallback: OpenRouter GPT Image 2.
Last resort: Grok or another model only when explicitly needed.

ForgeBot Change Made

The governing image-gen skill now documents the verified subscription route, its prerequisites, the command pattern that unsets API-key billing routes, output verification, and fallback behavior.

Bottom line: the OpenRouter 402 should not have forced a Grok downgrade. ForgeBot had a working GPT Image 2 subscription route available through Codex.

Evidence Table

Source	Date	Claim	Confidence
OpenAI Codex issue #8758	Apr. 16, 2026	OpenAI maintainer confirms image generation works when logged in with ChatGPT; issue closed completed.	High
Codex image-generation source	Current main	`IMAGE_MODEL = "gpt-image-2"`.	High
Codex feature registry	Current main	Image generation is stable and enabled by default.	High
OpenAI Codex authentication docs	Current	ChatGPT sign-in provides subscription access; API-key sign-in uses standard API pricing.	High
OpenAI Codex access-token docs	Current	Codex tokens are for Codex workflows; use Platform API keys for general OpenAI API calls.	High
OpenAI billing separation	Updated Jun. 2026	ChatGPT and OpenAI Platform API billing remain separate.	High
Controlled local execution	Jul. 13, 2026	ChatGPT-authenticated Codex generated a PNG with OpenAI/OpenRouter API-key variables removed.	High