Twenty AI models and agents from OpenAI, Google and Mistral are due to switch off in October 2026, and two Mistral models were due to retire on 30 September. If your app or script calls any model ID on this page, change it before the date, or it will stop working. Each row comes from the vendor’s own deprecations page or changelog, re-read on 13 October 2026. One more, the original Nano Banana, was due on 2 October until Google moved it to 15 March 2027.1
- 1 Oct: OpenAI’s
gpt-5.4-cyber. - Moved: the original Nano Banana,
gemini-2.5-flash-image, was due on 2 Oct on Google’s Gemini API; Google now gives 15 Mar 2027.1 - 5 Oct: Google’s old Antigravity agent,
antigravity-preview-05-2026. - 22 Oct (date passed): Google’s three Veo 3.1 video previews on the Gemini API, which Google now says were shut down, and the Gemini Omni preview, whose date has passed with no separate note from Google (checked 3 Nov 2026).
- 23 Oct: a batch of older OpenAI models, including
gpt-4-turbo,gpt-4-0613,o1-2024-12-17ando3-mini, and Google’s old Deep Research agent. - 31 Oct: Z.ai GLM 5.2 on Mistral’s API.
- Date passed, 30 Sep: Mistral’s OCR 4.0 and Leanstral 1.5.
- Next month, 2 Nov: xAI’s
grok-imagine-image-quality. - If you only use ChatGPT, Gemini or Grok in an app, none of this needs action from you. These are developer model IDs.
Deprecated means the company has announced that a model is on its way out. It usually still works for now. Shut down (or retired) means it is actually switched off: requests to it fail. A model ID is the exact name a program sends, like gpt-4-turbo. An endpoint is the address a program sends its requests to. A preview is an early test version that the company can change or end.
Google puts it this way: “A ‘deprecation’ is the announcement that we no longer provide support for a model, and that it will be ‘shut down’ in the near future. Once a model is ‘shutdown’, it is completely turned off, and the endpoint is no longer available.”2 OpenAI says the same in its own words: “At the time of the shut down, the model or endpoint will no longer be accessible.”3
The calendar
One row per switch-off, in date order. “Use instead” is what the vendor itself recommends.
| Date | What stops | Who it affects | Use instead | Source |
|---|---|---|---|---|
| 30 Sep (date passed) | OCR 4.0, mistral-ocr-4-0 | Developers using Mistral’s model for reading text from scanned documents | mistral-ocr-4-1 or mistral-ocr-latest | Mistral changelog |
| 30 Sep (date passed) | Leanstral 1.5, labs-leanstral-1-5 | Developers using Mistral’s experimental model for Lean 4 maths proofs | None named | Mistral changelog |
| 1 Oct | gpt-5.4-cyber | Developers using OpenAI’s cyber model through the API | “The most capable cyber model available to you” | OpenAI deprecations |
| 2 Oct, now moved to 15 Mar 2027 | Original Nano Banana, gemini-2.5-flash-image, on the Gemini API | Developers making images with the 2025 Nano Banana model | Nano Banana 2 Lite, gemini-3.1-flash-lite-image | Google deprecations |
| 5 Oct | Old Antigravity agent, antigravity-preview-05-2026 | Developers calling Google’s managed coding agent through the Gemini API | antigravity-preview-09-2026 | Google deprecations |
| 22 Oct | Veo 3.1 previews: veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview | Developers making video with Veo 3.1 on the Gemini API | gemini-omni-1.1-flash | Google deprecations |
| 22 Oct | Gemini Omni preview endpoint, gemini-omni-flash-preview | Developers who made video with the Omni preview on the Gemini API | gemini-omni-1.1-flash | Google changelog and deprecations |
| 23 Oct | Twelve older OpenAI model snapshots (full list below) | Developers still calling GPT-3.5, GPT-4, GPT-4 Turbo, o1, o3-mini, o4-mini and others by those IDs | GPT-5.6 Sol, Terra or Luna; GPT Image 2.5 for images | OpenAI deprecations |
| 23 Oct | Old Deep Research agent, deep-research-pro-preview-12-2025 | Developers calling Google’s Deep Research agent through the Gemini API | deep-research-preview-04-2026 | Google release notes and deprecations |
| 31 Oct | Z.ai GLM 5.2, zai-glm-5-2 | Developers using that Z.ai model through Mistral’s API | zai-glm-5-3 | Mistral changelog |
| 2 Nov (next month) | grok-imagine-image-quality | Developers using that xAI image model name | grok-imagine-image-2.0 with a quality setting | xAI migration guide |
Google, Gemini API deprecations, last updated 12 Oct 2026, and Gemini API changelog · Mistral, Changelogs · OpenAI, Deprecations · xAI, grok-imagine-image-quality migration guide. All read at source 13 Oct 2026; quotes are in each section below.
Search your code and settings for every model ID in the “What stops” column. A match means a job that will start failing on that date.
Treat Google’s dates as the earliest a model can go, not a promise: its deprecations page says “The shutdown dates listed in the table indicate the earliest possible dates on which a model might be retired.”4
Date passed: 30 September
OCR 4.0. Mistral’s OCR 4.0 (mistral-ocr-4-0), a model that reads the text in scanned pages and images, was due to retire on 30 Sep 2026. Mistral names OCR 4.1 (mistral-ocr-4-1 or mistral-ocr-latest) as the replacement, and has not posted a note confirming the retirement.5First-hand: until 13 Oct 2026 this page left out OCR 4.0 and Z.ai GLM 5.2, both in the same Mistral changelog entry as Leanstral 1.5, and listed the Gemini Omni preview here as ended on 30 Sep. Google now gives the Omni preview a 22 Oct date (see below).
Leanstral 1.5. Mistral released it on 30 Jun 2026 as an experimental “Labs” model for Lean 4, a language for writing mathematical proofs that a computer can check. It said from the start that the model would retire on 30 Sep 2026, and repeated that on 29 Sep. Mistral names no replacement, and as of 13 Oct 2026 it has not posted a note confirming the retirement.6
October, date by date
1 October: gpt-5.4-cyber
OpenAI scheduled its gpt-5.4-cyber model to shut down on 1 Oct 2026. It announced this on 11 Sep. OpenAI does not name one replacement; it tells developers to move to “the most capable cyber model available to you”. Re-read on 13 Oct 2026, the entry still gives the date with no note confirming the shutdown.7
Moved to March 2027: the original Nano Banana
Google’s first Nano Banana image model, gemini-2.5-flash-image, no longer shuts down in October. Google’s Gemini API deprecations page now gives 15 March 2027, and names Nano Banana 2 Lite (gemini-3.1-flash-lite-image) as the replacement.1 On 1 October the same page gave 2 October 2026, a year after the model’s release.8
Nano Banana 2 (gemini-3.1-flash-image), which this page suggested before, is itself now deprecated, with no shutdown date announced. Google recommends Nano Banana 2.1 (gemini-nano-banana-2.1) for new projects.910
The Gemini API date now matches Google Cloud’s Agent Platform (formerly Vertex AI), which moved the same model to 15 Mar 2027 on 14 Sep 2026. How to prompt the newer models is on the Nano Banana cheat sheet.11
5 October: the old Antigravity agent
Google’s managed coding agent in the Gemini API, antigravity-preview-05-2026, was due to shut down on 5 Oct 2026. Google’s pages still give that date, with no note confirming the shutdown. Its replacement, antigravity-preview-09-2026, came out on 17 Sep. This is the agent that developers call from their own code. It is not the Antigravity desktop app or the Antigravity CLI.4 Switching is not just a name change: the new version renames its tool settings. The detail is on Gemini CLI to Antigravity.
22 October: Google’s Veo 3.1 and Omni previews
Google says its three Veo 3.1 video previews on the Gemini API were shut down on 22 Oct 2026. This page had them as scheduled on 13 Oct 2026; Google’s Gemini API video page, read on 3 Nov 2026, now says they were shut down.12 They are veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, and Google names gemini-omni-1.1-flash, its Gemini Omni Flash video model, as the replacement for all three.4 This is the Gemini API only: on Google Cloud’s Gemini Enterprise Agent Platform, Veo 3.1 and Veo 3.1 Fast are generally available, with a retirement date of “November 17, 2026 or later”.13
The Gemini Omni preview had the same date. Google made Gemini Omni Flash fully available as gemini-omni-1.1-flash on 27 Aug 2026 and first said the old preview address would end on 30 Sep.14 Its deprecations page lists gemini-omni-flash-preview under Deprecated models with a shutdown date of 22 Oct 2026, with gemini-omni-1.1-flash as the replacement.4 That date has passed, and as of 3 Nov 2026 Google has not posted a separate note saying this preview is off.15 What the new version can do is on Veo 3.1 vs Gemini Omni vs Grok Imagine Video.
23 October: OpenAI’s older models
On 23 Oct 2026 OpenAI plans to shut down twelve older model snapshots. A snapshot is a model frozen at one date, such as o1-2024-12-17; the shorter names after the bar are aliases that point to it and stop with it. OpenAI announced this on 22 Apr 2026. This is the full list of snapshots as OpenAI gives it, with OpenAI’s own substitutes.
| Model snapshot (and aliases) | OpenAI’s substitute |
|---|---|
gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions | gpt-5.6-terra |
gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions | gpt-5.6-sol |
gpt-4-1106-preview | gpt-5.6-sol |
gpt-4-turbo | gpt-4-turbo-2024-04-09, gpt-4-turbo-completions | gpt-5.6-sol |
gpt-4.1-nano | gpt-4.1-nano-2025-04-14 | gpt-5.6-luna |
gpt-4o-2024-05-13 | gpt-5.6-sol |
gpt-image-1 | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
o1-2024-12-17 | o1 | gpt-5.6-sol |
o1-pro-2025-03-19 | o1-pro | gpt-5.6-sol (reasoning.mode: pro) |
o3-mini-2025-01-31 | o3-mini | gpt-5.6-sol |
ft-o4-mini-2025-04-16 | gpt-5.6-terra |
o4-mini-2025-04-16 | o4-mini | gpt-5.6-terra |
OpenAI, Deprecations, “2026-04-22: Legacy GPT model snapshots” entry, read at source 13 Oct 2026: “Access to these models will be shut down on the dates below.” Each row in its table gives “October 23, 2026”.
The same entry also removes five fine-tuned versions that day: ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano-2025-04-14, ft-babbage-002 and ft-davinci-002. A fine-tuned version is a copy of a model that a developer trained further on their own examples.16
The substitutes are GPT-5.6 models, because that was OpenAI’s current family when it wrote the list in April. OpenAI has since released GPT-6 models; the deprecations page has not changed its substitutes. What OpenAI offers now is on OpenAI’s model lineup.17
OpenAI’s newer image models follow on 1 Dec 2026. gpt-image-1.5, gpt-image-1-mini and chatgpt-image-latest shut down that day, again with GPT Image 2.5 (gpt-image-2.5-sunburst or gpt-image-2.5-flare) as the substitute. Re-read on 13 Oct 2026, the page still gives 23 Oct for gpt-image-1.17
23 October: Google’s old Deep Research agent
Google shuts down its deep-research-pro-preview-12-2025 agent on 23 Oct 2026. It announced this on 8 Oct and points developers to deep-research-preview-04-2026 (built for speed) or deep-research-max-preview-04-2026 (built for thoroughness). Like the Antigravity agent, this is an agent that developers call from their own code through the Gemini API.4
31 October: Z.ai GLM 5.2 on Mistral
Mistral retires Z.ai GLM 5.2 (zai-glm-5-2) from its API on 31 Oct 2026. GLM 5.2 is a model from the company Z.ai that Mistral also offers through its API. Mistral names Z.ai GLM 5.3 (zai-glm-5-3) as the replacement, “at the same price”.5
Next month: 2 November
xAI retires the model name grok-imagine-image-quality on 2 Nov 2026. It gave 60 days’ notice, from 2 Sep. Unusually, old requests will not simply fail: xAI says they will be answered by grok-imagine-image-2.0 at its lowest quality setting. To keep higher quality, switch to grok-imagine-image-2.0 and set quality yourself.18 The settings are on the Grok Imagine cheat sheet.
What to do this week
Five checks, in order.
- Search your code and config for each model ID above. Include environment files and no-code tools, where model names often hide.
- Swap in the vendor’s substitute and run a few real requests. A newer model can answer differently, so read the output, not just the error log.
- Check settings, not only names. Google’s new Antigravity agent renames its tool settings; xAI’s new image model needs a
qualityvalue. - Check the cost on the vendor’s own price page before you switch a busy job. This site no longer tracks prices: Google · xAI.
- Re-read the vendor page on the day. Dates can move: Google has moved the original Nano Banana’s shutdown on both the Agent Platform and the Gemini API, and the Gemini Omni preview from 30 Sep to 22 Oct.
Why vendors retire models at all, and how much notice they usually give, is on when AI models retire.
What this page could not verify
- Whether each switch-off actually happened. As of 13 Oct 2026, no vendor here has posted a note confirming that a shutdown on this page took place. This page reports scheduled dates from the vendors’ own pages. Update 3 Nov 2026: Google’s Gemini API video page now says the three Veo 3.1 previews were shut down on 22 Oct 2026 (see the 22 October section). The other dates on this page were not re-read on 3 Nov 2026.
- The Sora API is no longer in doubt. OpenAI shut it down on 24 Sep 2026: its Sora 2 model page now confirms it, though its deprecations page still listed the date with no note when re-read on 13 Oct 2026.19 The full story is on the Sora API shutdown page.20
- OpenAI’s 28 Sep batch (
gpt-3.5-turbo-instruct,babbage-002,davinci-002,gpt-3.5-turbo-1106). Scheduled for 28 Sep 2026; OpenAI has not confirmed it took place.3 - Anything not on a vendor’s own page. Rumoured retirements are left out on purpose.
- 1Google, Gemini deprecations, read at source 13 Oct 2026gemini-2.5-flash-image, shutdown “March 15, 2027”, replacement “gemini-3.1-flash-lite-image”. ↑agemini-2.5-flash-image, shutdown “March 15, 2027”, replacement “gemini-3.1-flash-lite-image”. ↑bgemini-2.5-flash-image, shutdown “March 15, 2027”, replacement “gemini-3.1-flash-lite-image”. ↑c
- 2Google, Gemini API deprecations, last updated 12 Oct 2026, read at source 13 Oct 2026.↑
- 3OpenAI, Deprecations, read at source 13 Oct 2026.↑a↑b
- 4Google, Gemini API deprecations, last updated 12 Oct 2026, read at source 13 Oct 2026“The shutdown dates listed in the table indicate the earliest possible dates on which a model might be retired.” ↑aantigravity-preview-05-2026, released “May 19, 2026”, shutdown “October 5, 2026”, replacement antigravity-preview-09-2026 · Gemini API release notes, 17 Sep 2026 entry, read at source 13 Oct 2026: “antigravity-preview-05-2026 shuts down on October 5, 2026, tracked on the deprecations page.” ↑bveo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, shutdown “October 22, 2026”, recommended replacement “gemini-omni-1.1-flash”; the page notes that its dates “indicate the earliest possible dates on which a model might be retired”. ↑cgemini-omni-flash-preview, released “June 30, 2026”, shutdown “October 22, 2026”, replacement “gemini-omni-1.1-flash”; gemini-omni-1.1-flash, “No shutdown date announced”. ↑ddeep-research-pro-preview-12-2025, shutdown “October 23, 2026”, replacement “deep-research-preview-04-2026” · Gemini API release notes, 8 Oct 2026 entry, read at source 13 Oct 2026: “The deep-research-pro-preview-12-2025 agent is deprecated and will be shut down on October 23, 2026.” ↑e
- 5Mistral, Changelogs, 29 Sep 2026 entry, read at source 13 Oct 2026“OCR 4.0 (mistral-ocr-4-0) is deprecated and retires on September 30, 2026. Use OCR 4.1 (mistral-ocr-4-1 or mistral-ocr-latest) instead” · “Z.ai GLM 5.2 (zai-glm-5-2) is deprecated and retires on October 31, 2026. Use Z.ai GLM 5.3 (zai-glm-5-3) instead”. ↑a“OCR 4.0 (mistral-ocr-4-0) is deprecated and retires on September 30, 2026. Use OCR 4.1 (mistral-ocr-4-1 or mistral-ocr-latest) instead” · “Z.ai GLM 5.2 (zai-glm-5-2) is deprecated and retires on October 31, 2026. Use Z.ai GLM 5.3 (zai-glm-5-3) instead”. ↑b
- 6Mistral, Changelogs, read at source 13 Oct 2026, 29 Sep 2026 entry“Leanstral 1.5 (
labs-leanstral-1-5), an experimental Labs model, is deprecated and retires on September 30, 2026.” ↑ - 7OpenAI, Deprecations, 2026-09-11 entry, read at source 13 Oct 2026“The gpt-5.4-cyber model is deprecated and will be removed from the API on October 1, 2026. Migrate to the most capable cyber model available to you before the shutdown date.” ↑
- 8Google, Gemini API deprecations, last updated 30 Sep 2026, read at source 1 Oct 2026
gemini-2.5-flash-image, released “October 2, 2025”, shutdown “October 2, 2026”. ↑ - 9Google, Gemini API release notes, 6 Oct 2026, read at source 13 Oct 2026“The gemini-3.1-flash-image model is deprecated (no shutdown date announced). Migrate to gemini-nano-banana-2.1.” ↑
- 10Google, Nano Banana image generation, read at source 13 Oct 2026“The previous-generation high-efficiency workhorse model, balancing speed with 4K generation, world knowledge, and multi-reference image consistency. Recommended for all new projects to use Nano Banana 2.1.” ↑
- 11Google Cloud, Agent Platform release notes, 14 Sep 2026 entry, read at source 13 Oct 2026“Deprecated and scheduled for retirement on March 15, 2027 (extended from October 2, 2026).” ↑
- 12Google, Video generation in the Gemini API, read at source 3 Nov 2026“Veo 3.1 Preview (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview) in the Gemini API was shut down on October 22, 2026; migrate to Gemini Omni Flash (gemini-omni-1.1-flash) or the GA models available through the Gemini Enterprise Agent Platform to avoid service interruptions.” ↑
- 13Google Cloud, Veo 3.1, Gemini Enterprise Agent Platform documentation, last updated 10 Oct 2026, read at source 13 Oct 2026veo-3.1-generate-001 and veo-3.1-fast-generate-001, “Launch stage: GA”, retirement date “November 17, 2026 or later”; veo-3.1-lite-generate-001, “Launch stage: Preview”. ↑
- 14Google, Gemini API release notes, 27 Aug 2026 entry, read at source 13 Oct 2026“Released gemini-omni-1.1-flash, the GA version of our fast, conversational video generation and editing model.” · “The existing gemini-omni-flash-preview endpoint will be deprecated on September 30, 2026.” ↑
- 15Google, Gemini API deprecations, read at source 3 Nov 2026under “Deprecated models”, gemini-omni-flash-preview, released “June 30, 2026”, shutdown “October 22, 2026”, replacement “gemini-omni-1.1-flash”; no separate note on that page or on Google’s video page says this preview has been shut down. ↑
- 16OpenAI, Deprecations, 2026-04-22 entry, read at source 13 Oct 2026“We are also removing fine-tuned versions as below”: ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano-2025-04-14, ft-babbage-002 and ft-davinci-002, each “October 23, 2026”. ↑
- 17OpenAI, Deprecations, read at source 13 Oct 2026substitute column as in the table above · OpenAI, API changelog, read at source 13 Oct 2026: GPT-6 Astra released 3 Sep 2026, GPT-6 Sol and Luna 22 Sep 2026. ↑a“On June 2, 2026, we notified developers using older GPT Image models of their deprecation and removal from the API on December 1, 2026.” The table lists gpt-image-1.5, gpt-image-1-mini and chatgpt-image-latest for Dec 1, 2026, and gpt-image-1 for October 23, 2026, each with the replacement “gpt-image-2.5-sunburst or gpt-image-2.5-flare”. ↑b
- 18xAI, grok-imagine-image-quality migration guide, read at source 13 Oct 2026“Requests to
grok-imagine-image-qualityare served bygrok-imagine-image-2.0withqualityset tolow”; retirement date “November 2, 2026”. ↑ - 19OpenAI, Sora 2 model page, read at source 13 Oct 2026“The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available.”; “There is no one-to-one replacement API.” ↑
- 20OpenAI, Deprecations, 2026-03-24 entry, read at source 13 Oct 2026removal “from the API on September 24, 2026”. ↑
Google, Gemini API deprecations (last updated 30 Sep 2026) and Gemini API changelog · OpenAI, Deprecations · Mistral, Changelogs · xAI, grok-imagine-image-quality migration guide. All read at source on 1 October 2026. Also: Google, Gemini Developer API pricing, read in a browser 18 Sep 2026 · Google Cloud, Agent Platform release notes, read in a browser 22 Sep 2026 · OpenAI, API changelog, read at source 29 Sep 2026.
Re-read at source on 13 October 2026, for a full re-check of this page: every source above, plus Google’s image generation page, the Google Cloud Veo 3.1 page and the Mistral changelog. The Google and OpenAI price pages are linked only, not quoted.
Shutdown dates move. Re-read the vendor page before you plan around one.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK