The 'gpt-5.5' option has appeared in the model selection dropdown of the OpenAI Codex interface, but is not yet available to ChatGPT users
Some developers have noticed that the model selection dropdown in the OpenAI Codex interface now includes the 'gpt-5.5' option, which is listed at the top. However, when attempting to select this model and submit a request under a ChatGPT account, a 400 error is immediately returned, indicating that "The 'gpt-5.5' model is not supported when using Codex with a ChatGPT account." This suggests that the model is currently unavailable or unauthorized for front-end users. As of now, there has been no official announcement or technical documentation regarding GPT-5.5 on OpenAI's official website, developer blog, or model list page. Most discussions are coming from the Codex user community, GitHub issues, and social media screenshots, which are generally interpreted as an early exposure of backend configuration or a "shadow launch" test rather than an official release.
Source: Public information
ABAB AI Insight
From OpenAI's past patterns, this situation of "model names appearing first in Codex and internal interfaces before public release" is not unprecedented. The GPT-5.x series has been exposed multiple times through error messages or model lists in Codex and CLI before formal announcements. Currently, GPT-5.5 is visible only in the UI but is strictly denied for ChatGPT accounts, likely indicating that internal or restricted testing is already taking place in some controlled environments or pure API channels, but capacity, alignment, and security validation for broader developers and front-end users have not yet been completed.
For developers and the ecosystem, what truly matters is not the name "5.5," but the signal that OpenAI continues to adopt a "0.5 generation rhythm" in its product line: rapidly iterating reasoning quality, tool capabilities, and cost structures between major versions, while deeply binding with vertical workflows like Codex. The appearance of GPT-5.5 at the top of the Codex model list suggests that its priority deployment scenarios are likely still in code, agents, and automation workflows—essentially the "AI productivity layer" that OpenAI currently aims to firmly grasp, rather than merely replacing general chat models.
From an industry structure perspective, this kind of "unannounced appearance" reflects a new phase in the three-way game of "safety and alignment—commercial competition—operational stability" for cutting-edge models: on one hand, there is a need to continuously refresh capability boundaries in competition with entities like Anthropic and DeepSeek, while on the other hand, after safety incidents (such as recent Codex command injection and ChatGPT data leakage vulnerabilities), the pace of opening new models must become more cautious. The result is this shadow launch form where the backend is already integrated, the frontend is restricted, and the official stance is to silently observe community reactions and operational data.
In the longer term, the early emergence of GPT-5.5 essentially confirms that OpenAI will continue along the path of "intensive sub-version iterations" within the framework of "a single flagship + multi-level specialized models," rather than waiting for a completely disruptive GPT-6 to switch entirely. For enterprises accustomed to restructuring workflows and cost structures around model versions, the more important capability in the future may not be betting on a single version, but rather quickly assessing the marginal changes in quality, cost, and safety from continuous small improvements like "5.4→5.5→5.6," and making their code, agents, and product links flexible for switching and rollback.