Skip to content
gentic.news — AI News Intelligence Platform

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

GPT-5.5 Spotted in Unofficial Codex API Backdoor

GPT-5.5 Spotted in Unofficial Codex API Backdoor

Simon Willison reports GPT-5.5 is available via a Codex API backdoor, bypassing the official OpenAI API for image generation tasks, suggesting an unofficial release channel.

Share:

What Happened

Transitioning from Codex to GPT 3.5 - API - OpenAI Developer Community

Simon Willison, a prominent technologist, tweeted that GPT-5.5—a model not officially listed in the OpenAI API—is accessible through what he describes as an "apparently approved-of Codex API backdoor." He used this to generate pelican images, sharing results in both default and "xhigh" resolutions. The tweet includes links to the generated images, though the exact endpoint or method isn't detailed.

Context

OpenAI's API typically lists models like GPT-4, GPT-4 Turbo, and GPT-5 in its official documentation. The appearance of GPT-5.5 via the Codex API—a separate service for code generation—suggests either an internal testing channel or a deliberate but unannounced release. Codex, originally for code completion, now appears to support image generation, indicating a broader multimodal expansion.

This isn't the first time OpenAI has used unofficial channels. Previous models like GPT-4.5 were first spotted in early access or third-party integrations before official launch. The "approved-of" qualifier implies OpenAI is aware of this access, possibly as a limited preview for developers.

What This Means in Practice

DCAgent/tbench-easy-unofficial-codex-gpt5-trials at main

For developers, GPT-5.5 via Codex offers a sneak peek at capabilities not yet in the main API. However, reliance on unofficial endpoints risks instability or sudden changes. The image generation feature—pelicans in this case—hints at multimodal improvements, but without benchmarks or official documentation, performance remains speculative.

Frequently Asked Questions

Is GPT-5.5 officially available?

No, GPT-5.5 is not listed in the official OpenAI API documentation. It's accessible only through the Codex API, which Simon Willison calls a "backdoor." OpenAI has not announced its release.

What can GPT-5.5 do that GPT-5 cannot?

Based on this report, GPT-5.5 appears to support image generation through the Codex API, a capability not publicly documented for GPT-5. The "xhigh" resolution option suggests enhanced output quality.

How can I access GPT-5.5?

Access requires using the Codex API endpoint, likely with specific parameters. Simon Willison's tweet doesn't provide exact instructions, but developers familiar with Codex might replicate his approach. Note that this is unofficial and may violate OpenAI's terms.

Will OpenAI officially release GPT-5.5?

Unclear. The "approved-of" language suggests OpenAI is aware of this access, possibly as a limited preview. An official release may follow if testing proves successful, but no timeline is available.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

The appearance of GPT-5.5 via the Codex API is notable for two reasons. First, it underscores OpenAI's ongoing experimentation with model versioning—GPT-5.5 suggests a minor iteration rather than a full point release, likely focused on specific improvements like image generation quality or latency. Second, using Codex as a delivery mechanism hints at a broader strategy to unify code and multimodal capabilities under a single API umbrella. For practitioners, this means monitoring unofficial channels for early signals of future API changes, but also caution: undocumented endpoints may lack reliability guarantees. From a competitive standpoint, this positions GPT-5.5 against models like Claude 3.5 Sonnet and Gemini 1.5 Pro, which already offer multimodal generation. If GPT-5.5 delivers superior image quality or lower latency, it could pressure rivals. However, without benchmark data or pricing, it's too early to assess real-world impact. The pelican examples—while amusing—are not rigorous evaluation. The "backdoor" access model also raises questions about OpenAI's release strategy. By offering limited access through Codex, they gather feedback without the scrutiny of a full launch. This mirrors past tactics with GPT-4.5, which appeared in ChatGPT before the API. Expect more leaks as OpenAI tests features internally.

Mentioned in this article

Enjoyed this article?
Share:

Related Articles

More in Products & Launches

View all