OpenAI is discontinuing GPT-5.5 from its ChatGPT consumer product and Codex on October 14, 2026, but the model will remain available through the OpenAI API platform for developers using API keys. This split retirement has created confusion in coverage, with many outlets flattening the announcement into a single shutdown date. The reality is more nuanced: which surface you use determines whether you need to migrate immediately or can continue working unchanged.

What's Actually Ending on October 14?

The October 14 retirement applies specifically to ChatGPT, ChatGPT Work, and Codex when accessed through a ChatGPT account sign-in across all subscription tiers, including consumer, Enterprise, and Edu plans. Users on these surfaces will lose access to GPT-5.5 and will need to migrate to GPT-5.6 Sol or GPT-6 Astra before the deadline. OpenAI has pointed Codex users toward GPT-5.6 Sol as the recommended replacement.

However, GPT-5.5 does not appear on OpenAI's official API deprecations page, which lists models scheduled for actual shutdown. This absence is significant: it means the API path survives the October 14 change. Developers authenticating with API keys rather than ChatGPT credentials can continue using GPT-5.5 through the API platform without interruption.

Why Does the Authentication Method Matter?

Codex, OpenAI's code-focused tool, exemplifies the confusion because it offers two ways to log in. Sign in with your ChatGPT account and you're drawing from your subscription plan capacity; sign in with an API key and you're drawing from your API credit balance. These are two different billing and access systems, and they're retiring on different timelines.

This pattern mirrors earlier retirements in 2026. When GPT-5.2 and GPT-5.2-Codex were sunset earlier this year, ChatGPT-authenticated users lost access on June 2, while API-key users retained access until later shutdown dates in July and August. The subscription surface always moves first.

What Should Developers Do Before October 14?

• Identify Your Access Path: Determine whether you authenticate through ChatGPT sign-in or API key. If you use "codex login --with-api-key," you're on the surviving path and can defer action.
• Audit Hardcoded Model IDs: If you're on the ChatGPT sign-in path, find every place where GPT-5.5 is hardcoded into your configuration, workspace defaults, saved settings, custom agents, or scheduled tasks. This is a search-and-replace problem that should be completed this week rather than waiting until October 13.
• Test Proxy Paths: If you reach GPT-5.5 through a gateway or proxy rather than directly, test that path before October 14. OpenAI tightened the Codex responses-lite endpoint in late June, and some proxies have been receiving errors for GPT-5.5 since then.

What Are GPT-5.5's Current Specifications?

For developers deciding whether to keep GPT-5.5 on the API platform, the model's terms remain unchanged by the announcement. GPT-5.5 is priced at $5.00 per million input tokens and $30.00 per million output tokens, with cached input reads at $0.50 per million tokens. The Pro tier costs $30.00 and $180.00 per million tokens respectively. The model supports a context window of roughly 922,000 input tokens and 128,000 output tokens, accepting text, image, and file input.

Performance metrics show GPT-5.5 scoring 39 on Artificial Analysis's Intelligence Index v4.3, with throughput around 94 output tokens per second and a time-to-first-token of approximately 35 seconds. OpenAI's own benchmarks report Terminal-Bench 2.0 at 82.7%, SWE-bench Verified at 82.6%, and SWE-bench Pro at 58.6%, though these figures remain vendor-reported and have not been independently verified by neutral laboratories.

Why Is Independent Evaluation Signaling Retirement?

Artificial Analysis, a neutral AI evaluation platform, has already flagged GPT-5.5 as deprecated on its leaderboard and reduced its testing workloads to default 10,000-token benchmarks, a signal typically associated with product retirement. This independent cooling of evaluation is worth monitoring because once neutral testing stops, the remaining data points become increasingly vendor-controlled.

The practical lesson for anyone building on OpenAI's platform is to read the specific surface, not the headline. "Discontinued" in news coverage can refer to the chat product, the API platform, or a specific authentication path, and those represent three entirely different migration scenarios.

When Should You Actually Migrate?

Two dates matter more than October 14. The first is whenever OpenAI officially adds GPT-5.5 to its API deprecations page; that announcement, not today's news, will signal the true end of the API path and will come with a formal notice period. The second is whether Artificial Analysis restores full benchmarking or retires the model entirely from evaluation, because independent assessment is what maintains confidence in a model's capabilities over time.

For teams using routing layers or API gateways that abstract away specific model IDs, the migration is minimal. A single configuration change pointing traffic to GPT-5.6 Sol or GPT-6 Astra can be tested and rolled out gradually, allowing you to measure performance on your actual workload rather than trusting vendor claims about the transition.