AI Model Retirements Q4 2026: What to Migrate and When

Update (20 October 2026): the 19 October GitHub Copilot removals have now taken effect, and GitHub’s final notice suggests newer replacements than its September announcement; the table and the Copilot section are updated.

Update (16 October 2026): GitHub has announced that Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra leave every GitHub Copilot experience on 16 November, and we have added its earlier notice for 19 October, which removes GPT-5.5, GPT-5.4 and four other models from Copilot. Both are in the table, a new GitHub Copilot section and checklist step 3.

Update (15 October 2026), correction: An earlier version of this article said GPT-5.5 was retiring from ChatGPT, ChatGPT Work and Codex on every plan, with GPT-6 Sol as the paid-plan replacement. That followed OpenAI’s documentation as worded on 11 October. OpenAI’s changelog entry of 14 October and its current models page scope the retirement to Codex for users signed in with ChatGPT, and name GPT-6.1 Sol as the replacement on paid plans. We have corrected the FAQ, the 14 October row, the ChatGPT section and checklist step 2.

Update (14 October 2026): OpenAI has added a new deprecation notice: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots leave the API on 14 January 2027. We have also added the OpenAI API shutdown on 23 October, announced in April, which removes gpt-3.5-turbo, gpt-4, o1, o3-mini, o4-mini and gpt-image-1, among others.

A string of AI model retirements lands between 14 October 2026 and 1 April 2027, and the next hard deadline is close: 23 October, when OpenAI switches off a batch of legacy API models, including GPT-4, GPT-3.5 Turbo and o1. GPT-5.5 has already gone from Codex for users signed in with ChatGPT, and on 19 October GitHub Copilot dropped it along with five other models. If you have saved settings, custom agents, scheduled tasks, no-code automations or API code pinned to an older model, this is the list to check. Every date below comes from the vendor’s own deprecation page or changelog.

The retirement calendar

Date	What retires	Replacement	Source
14 Oct 2026	GPT-5.5 in Codex for users signed in with ChatGPT (API and API-key Codex unaffected)	GPT-6.1 Sol (paid plans), GPT-6 Luna (Free, Go)	OpenAI changelog
23 Oct 2026	OpenAI API gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4-1106-preview, gpt-4o-2024-05-13, o1, o1-pro, o3-mini	gpt-5.6-terra (GPT-3.5 Turbo), gpt-5.6-sol (the rest; reasoning.mode: pro for o1-pro)	OpenAI deprecations
23 Oct 2026	OpenAI API gpt-4.1-nano, o4-mini, gpt-image-1	gpt-5.6-luna, gpt-5.6-terra, gpt-image-2.5-sunburst or gpt-image-2.5-flare	OpenAI deprecations
19 Oct 2026	GitHub Copilot: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Grok 4.5	Gemini 3.8 Flash, GPT-6.1 Sol (GPT-5.5, GPT-5.4), GPT-6 Luna (the minis), Grok 4.7	GitHub changelog
23 Oct 2026	Gemini API Deep Research agent deep-research-pro-preview-12-2025	deep-research-preview-04-2026 or deep-research-max-preview-04-2026	Gemini API changelog
24 Oct 2026	Perplexity Agent and Orchestrator API: openai/gpt-5.4 (incl. mini, nano), gpt-5.2, gpt-5.1, gpt-5, gpt-5-mini	A current OpenAI model from Perplexity’s model list	Perplexity API changelog
Nov 2026	Gemini Gems on personal accounts transition to skills; on work accounts Gems move into Settings on 17 Nov (still usable)	Skills in the Gemini app	Gemini Apps Help, Google Workspace Updates
16 Nov 2026	GitHub Copilot: Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra (Opus 4.8 stays for individual subscribers on annual plans)	Claude Opus 5.5; GPT-6.1 Sol or GPT-6 Luna	GitHub changelog
30 Nov 2026	Claude API claude-sonnet-4-5-20250929	claude-sonnet-5-5	Anthropic model deprecations
6 Jan 2027	OpenAI API tts-1, tts-1-hd, gpt-4o-mini-tts-2025-03-20, gpt-4o-mini-tts-2025-12-15	gpt-realtime-2.1-mini	OpenAI deprecations
14 Jan 2027	OpenAI API gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, gpt-4.1-mini-2025-04-14 (and their fine-tunes)	gpt-6-luna	OpenAI deprecations
1 Apr 2027	OpenAI API gpt-5.1, gpt-5.3-codex, gpt-5.4-nano	gpt-6-sol (first two), gpt-6-luna	OpenAI deprecations

ChatGPT users: GPT-5.5 left Codex on 14 October

This one affects people, not just developers, but it is narrower than first announced. According to OpenAI’s changelog, GPT-5.5 retired at 19:00 UTC on 14 October from Codex for users signed in with their ChatGPT account. The OpenAI API and Codex sessions authenticated with an API key keep it. OpenAI’s replacement is GPT-6.1 Sol (gpt-6.1-sol) on Plus, Pro, Business, Enterprise and Edu, and GPT-6 Luna in the desktop app on Free and Go. Admins should replace gpt-5.5 in Codex workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks and any script that selects a model. If you have not looked at ChatGPT’s newer models yet, our ChatGPT review covers what each plan gets now.

Gemini users: Gems become skills, slowly

Google is replacing Gems with skills: reusable instructions that run inline and can be stacked in one prompt. On personal Google accounts, Google’s help centre says Gems transition to skills starting in November 2026, and existing Gems are recreated as skills automatically when they go away. Work and school accounts get more time: skills started rolling out in the Gemini app on 13 October, Gems move to the Settings panel on 17 November and keep working, and the hard stop is no sooner than 1 March 2027 for business and enterprise (1 June 2027 for education), when remaining Gems are auto-migrated to draft skills. Skills do not yet support every Gem feature; Canvas and Deep Research, for example, do not work with them yet. One catch: skills do not sync between the Gemini app and Workspace apps, so a skill you want in both has to be created twice. Our Gemini skills guide has the full timeline and the steps to move a Gem by hand, and our Gemini review has the wider picture.

GitHub Copilot users: two rounds of model removals

GitHub is pruning Copilot’s model picker twice this autumn, and both rounds apply to every Copilot experience: Chat, inline edits, ask and agent modes, and code completions. On 19 October, Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5 went, and GitHub’s 19 October notice names GPT-6.1 Sol (for GPT-5.5 and GPT-5.4), GPT-6 Luna (for the two minis), Gemini 3.8 Flash and Grok 4.7 as the replacements, newer than the ones in its September announcement. On 16 November, announced on 15 October, Claude Opus 4.8 and Claude Opus 5 go (GitHub suggests Claude Opus 5.5), along with GPT-5.6 Terra (GPT-6.1 Sol or GPT-6 Luna). The exception: individual subscribers on annual plans keep Claude Opus 4.8.

For Copilot Business and Enterprise, the alternatives are switched on automatically under default model enablement, unless an admin has turned off the global default or disabled a specific model; in that case, enable them in the model policies in Copilot settings. The retired models disappear on their own, so the real work is updating custom agents, prompt files, workflows and integrations that name a model explicitly. Because Copilot now bills by tokens, check the replacement’s price too: our AI coding assistants guide covers Copilot’s plans and pricing.

Developers: the API deadlines to calendar

OpenAI legacy models (23 October). The most urgent one, announced in April. Besides the models in the table (including their completions variants), fine-tunes built on GPT-3.5 Turbo, GPT-4, GPT-4.1 nano, o4-mini, babbage-002 and davinci-002 stop the same day. Old scripts and plugins often hard-code gpt-4 or gpt-3.5-turbo, so search for those strings first.

GPT-4o and GPT-4.1 (14 January 2027). Announced on 14 October, this one will reach more people, because GPT-4o mini and GPT-4.1 mini are still the default cheap models in many integrations and no-code automations. OpenAI names gpt-6-luna as the replacement for all five snapshots, and fine-tuned versions follow the same schedule. Test output quality and cost on Luna before you switch, rather than assuming a like-for-like swap. OpenAI’s page also lists December shutdowns for older GPT Image models (1 December) and the original GPT-5 and o3 snapshots (11 December). To check programmatically, the List models endpoint returns a shutdown_date field for each model.

Claude Sonnet 4.5 (30 November). Moving to Sonnet 5.5 is more than a string swap. Anthropic lists breaking changes, including errors for forced tool use and for non-default sampling parameters. Our Claude Sonnet 5.5 launch coverage walks through them. Note also that claude-opus-4-5-20251101 is listed as retiring “not sooner than” 24 November 2026; that is a floor, not an announced date.

OpenAI text-to-speech (6 January 2027). tts-1 still sits inside many voice apps and no-code automations. The replacement is a realtime model, so plan for a different API shape, not just a new model name. If you are reconsidering your voice stack entirely, see our best AI voice generators guide.

Perplexity and Gemini API (late October). Both are short notices. After 24 October Perplexity will reject the old GPT-5.x IDs outright, so check fallback chains too, not just primary model choices.

A ten-minute migration checklist

1. Search your code, configs and no-code automations for the exact model IDs in the table.
2. Check saved Codex defaults in ChatGPT workspace settings, custom agents and scheduled tasks, since GPT-5.5 is gone from Codex with ChatGPT sign-in as of 14 October (switch to GPT-6.1 Sol, or GPT-6 Luna on Free and Go).
3. In GitHub Copilot, replace explicit references to the models leaving on 19 October and 16 November (including Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra) in custom agents, prompt files and workflows, and check that Business or Enterprise model policies enable the replacements.
4. Re-run a few real prompts on the replacement model and compare output, length and cost.
5. Read the replacement’s breaking-changes notes (Sonnet 5.5 and the realtime TTS models have them).
6. Put the remaining dates in a shared calendar with an owner.

For a primer on reading vendor announcements like these, see how to read AI model announcements. More updates are in our news section.