# OpenAI Will Retire GPT-5.5 From ChatGPT and Codex on October 14 > Source: > Published: 2026-09-15 22:44:49+00:00 *OpenAI is pulling GPT-5.5 out of ChatGPT and Codex on October 14, and any founder with the model wired into a live pipeline needs to swap it out before then or watch the calls fail.* Mark the date. October 14 is when GPT-5.5 and GPT-5.5 Thinking stop answering inside ChatGPT and the Codex CLI. The news surfaced after developers on OpenAI's own community forum asked the company to keep GPT-5.5 around as a legacy option in Codex, the way older models sometimes linger for holdouts. OpenAI's answer was no. The model only shipped on April 23, 2026. By October 14 it will be dead before its seventh month is out. If you're a founder or engineer with GPT-5.5 hardcoded into an API call, a Codex CLI script, or a scheduled agent task, you have about a month. Miss the window and the calls don't slow down or degrade. They just fail. Any request that names the model directly comes back with an error instead of a response, the same failure mode developers ran into in June when OpenAI cut gpt-5.2 and gpt-5.3-codex loose from ChatGPT. That migration forced teams to comb through workspace defaults, saved model settings, custom agent configs, and every codex exec --model flag in their scripts. This one asks for the same work, on a tighter clock. ## The GPT-5.6 lineup taking its place OpenAI wants everyone moved onto GPT-5.6 instead, the family it introduced in July under three new names: Sol, Terra, and Luna. Sol carries the highest reasoning ceiling, built for chewing through large codebases and long agentic runs. Terra is the balanced, everyday default, and Codex already slots it in wherever GPT-5.4 used to sit. Luna is the cheap, fast option for small jobs. OpenAI prices the three at $5 input and $30 output per million tokens for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna. GitHub confirmed the same three models rolled into GitHub Copilot back in July, so this isn't a Codex-only shift. It's the naming scheme OpenAI now expects across its whole developer stack. Not everyone gets hit at once, either. Commercial developers calling GPT-5.5 straight through the API have more room than ChatGPT subscribers and Codex users signed in with a ChatGPT account, whose access simply ends on the 14th. That distinction matters if your product calls the model server-side versus routing requests through a logged-in Codex session, and it's worth checking which one your stack actually does before assuming you're covered. [A Codex reasoning leak claim puts AI tool trust back on the table](https://startupfortune.com/a-codex-reasoning-leak-claim-puts-ai-tool-trust-back-on-the-table/) A Reddit-driven claim says GPT-5.5 reasoning text may be leaking in Codex, but the evidence is not confirmed. The bigger issue is that startups now rely on AI coding agents deeply enough to need logging, redaction and vendor monitoring policies. - [how to prevent AI coding agent data leaks](https://startupfortune.com/a-codex-reasoning-leak-claim-puts-ai-tool-trust-back-on-the-table/) - [managing security risks with AI development tools](https://startupfortune.com/a-codex-reasoning-leak-claim-puts-ai-tool-trust-back-on-the-table/) ## A pattern worth sitting with Here's the pattern, and it's worth sitting with. GPT-5.5 launched April 23. GPT-5.6 arrived in July, about eleven weeks later, first as a limited preview for roughly 20 partner organizations before OpenAI widened access. By October 14, GPT-5.5 is retired entirely. Before it, GPT-5.2 and GPT-5.3-Codex got cut from ChatGPT on June 2. The gpt-5.4-cyber variant leaves the API on October 1, two weeks ahead of GPT-5.5 itself. Four retirements in under five months. Frankly, that's what building on top of a closed frontier lab looks like right now. It's not a foundation you can set and forget. OpenAI's stated policy promises six months of notice for generally available models, and technically GPT-5.5's October 14 date clears that bar. But six months from a model most teams didn't even adopt until weeks after launch feels a lot shorter in practice, and the Codex community thread makes clear plenty of developers are annoyed about it. Teams running open-weight models like Llama keep the weights on their own hardware indefinitely. Teams building on ChatGPT or Codex don't get that option. They get a calendar. OpenAI's Codex documentation already spells out the fix for anyone who postponed it during the gpt-5.4 wind-down: replace gpt-5.4 with gpt-5.6-terra and gpt-5.4-mini with gpt-5.6-luna everywhere a model name might be saved. GPT-5.5 users get the identical instruction now, with a harder deadline. October 14 is five weeks away. **Also read:** [Universal Music Group Sues DistroKid Over AI Slop Flooding Streaming](https://startupfortune.com/universal-music-group-sues-distrokid-over-ai-slop-flooding-streaming/) • [How to Test AI Agent Output Guardrails Before Shipping to Production](https://startupfortune.com/how-to-test-ai-agent-output-guardrails-before-shipping-to-production/) • [US data centers will burn more natural gas than almost every country by 2035](https://startupfortune.com/us-data-centers-will-burn-more-natural-gas-than-almost-every-country-by-2035/) *This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.* [OpenAI Brings Back Five-Hour Codex Limits After Six Weeks of Unlimited Access](https://startupfortune.com/openai-brings-back-five-hour-codex-limits-after-six-weeks-of-unlimited-access/) OpenAI reimposed its five-hour rolling usage limit on ChatGPT Plus Codex and Work environments on August 25, 2026, ending six weeks of uncapped access that began with the GPT-5.6 Sol rollout on July 13. The reversal follows two earlier quota resets in August and bug fixes for abuse-detection systems that had been draining users' allowances faster... - [openai restores five hour codex usage limits](https://startupfortune.com/openai-brings-back-five-hour-codex-limits-after-six-weeks-of-unlimited-access/) - [chatgpt plus codex access limits reimposed august](https://startupfortune.com/openai-brings-back-five-hour-codex-limits-after-six-weeks-of-unlimited-access/) ## Join the discussion [Open in the community →](https://startupfortune.com/community/) Almost there. Sign in and your reply posts straight away.