Schedule · checked 2026-10-19

October 2026 AI releases and retirements: dates the vendors published, wording they only hedged, and fields left blank

As of 2026-11-05, these further updates carry an official date: on 2026-10-26 OpenAI released GPT-6.1 Terra (gpt-6.1-terra) in the API, callable through Responses and Chat Completions; from 2026-10-31 Kimi's K2 series closed to new sign-ups (existing users are not affected) and Kimi plans to take all K2 models offline at the end of December; and on 2026-11-02 Google deprecated gemini-2.5-pro and gemini-2.5-flash; from 2026-12-01 all API requests to them will be routed automatically to gemini-3.8-flash and gemini-3.5-flash-lite respectively and billed at the new models' rates. As of 2026-10-19, these October dates are written down by the vendors themselves: GPT-5.5 leaves ChatGPT, ChatGPT Work and Codex on 2026-10-14 (the official text says the OpenAI API is not affected); 2026-10-15 is the «not sooner than» floor for Claude Haiku 4.5; GitHub Copilot drops this batch of models on 2026-10-19; 2026-10-22 is the earliest shutdown date Google lists for several Veo 3.1 previews and gemini-omni-flash-preview; OpenAI shuts down a batch of older models on 2026-10-23; GitHub Universe runs 2026-10-28–29; and another batch of Copilot models leaves in mid-November, on 2026-11-16. Already done: on 2026-10-02 Copilot removed four models on the day it announced it, and on 2026-10-06 Mistral opened a public preview of Mistral Large 4 while OpenAI released the Decisions API in beta and cut its API usage tiers from five to three; on 2026-10-13 Gemini 4 Argon went GA, and on 2026-10-17 DeepSeek released DeepSeek-V4.1-Flash-1017. Where vendors gave only wording, this schedule copies it without converting; fields they left blank stay blank.

Updated 2026-11-10

One key across Claude, GPT and Chinese models, billed per token.

Billed per token — live rates on /models
Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
Self-service: top up and activate instantly

Create account View pricing

#AI release calendar#2026-10-23 shutdown#Haiku 4.5 retirement floor#DevDay Exchanges

Four calendar lines that get misread

One

«Not sooner than» is not a shutdown date

Every retirement entry in Anthropic's table reads «Not sooner than», and Google's Gemini API deprecations table notes that its shutdown dates are «the earliest possible dates». Both promise only that nothing happens before that day, not that it happens on it. Treating 2026-10-15 for Haiku 4.5, or 2026-10-22 for Google's previews, as a fixed switch-off date is reading a sentence that does not exist.

Two

The 2026-10-01 notices put no GPT-6 model up for shutdown

On 2026-10-01 OpenAI's deprecations page gained two notices: gpt-5.3-codex, gpt-5.1 and gpt-5.4-nano leave the API on 2027-04-01 (replacements gpt-6-sol and gpt-6-luna), and text-to-speech models such as tts-1 and tts-1-hd leave on 2027-01-06 (replacement gpt-realtime-2.1-mini). GPT-6 models appear only in the replacement column, not on any shutdown list, so as of the check date the claim that the older GPT-6 models retire after DevDay still has no official basis.

Three

Eight Exchanges cities, no dates

The official DevDay page writes «Later this year, we're bringing DevDay Exchanges into Bengaluru, Tokyo, Seoul, Paris, Berlin, London, São Paulo, and Mexico City» — the cities are named, per-city dates and ticket prices are not, so this page lists cities and schedules none.

Four

The replacement column is the vendor's own choice, not ours

In the rows removed from the API on 2026-10-23 OpenAI fills the right-hand column itself: gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-image-2.5-sunburst or gpt-image-2.5-flare. gpt-5.6-luna is not currently supported on QCode, and neither is gpt-6-luna; use another GPT model for those requests, such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra. The vendor gives no order beyond that column and does not say which model fits your workload, so write new code along the right-hand column as written.

Why this page only carries these rows

A calendar page is only worth as much as its credibility: one official sentence per row. So we accept three kinds of item — a published date, a published time wording, and a published «not sooner than» floor. Two kinds we do not schedule: an outlet's «expected» and a founder's one-line post. Those are leads, not dates to plan a release around.

Already happened: 2026-10-01 to 2026-10-06, 2026-10-08 to 2026-10-17

2026-10-01: the API removal date OpenAI set for gpt-5.4-cyber arrived (the page reads «will be removed from the API on October 1, 2026»), and the same day two new deprecation notices were added (see above). 2026-10-02: GitHub removed Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code and Claude Opus 4.7 from all Copilot experiences as of that day; the suggested alternatives are Gemini 3.8 Flash for the first two, then Kimi K3 and Claude Opus 5.5. 2026-10-06: Mistral launched a public preview of Mistral Large 4 (model id mistral-large-4 in the docs, status Public Preview) and says the preview API can be tried on Mistral Studio from that day, with the weights following by the end of the month; the same day OpenAI released the Decisions API in beta with gpt-6-luna and simplified API usage tiers from five to three — Build, Launch and Grow — with organizations upgrading automatically as total credit purchases reach tier minimums. 2026-10-08: OpenAI's API changelog added Ultrafast mode for GPT-6.1 Sol in the Responses API. 2026-10-13: Google's release notes write Gemini 4 Argon (gemini-4-argon) generally available (GA), and the vendor calls it «our most capable model built for complex coding, agentic, math, and science». 2026-10-14: the day written on OpenAI's Models page arrived, GPT-5.5 retired from ChatGPT, ChatGPT Work and Codex on all plans, and the official text still notes that the OpenAI API is not affected. 2026-10-17: DeepSeek released DeepSeek-V4.1-Flash-1017, the release post is titled «Wish-Driven Programming Made Real», the vendor calls it «a Pareto improvement in performance and cost», and it is now live on the website, the app and the API.

The next 45 days

2026-10-19

Today (2026-10-19) is the day GitHub Copilot drops this batch of models, and the announcement reads «We will deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on October 19th, 2026:»; the models are Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. The row-by-row alternatives are Gemini 3.7 Flash → Gemini 3.8 Flash, GPT-5.5 → GPT-5.6 Sol, GPT-5.4 → GPT-5.4, GPT-5.4 mini → GPT-5.6 Luna, GPT-5 mini → GPT-5.6 Luna, Grok 4.5 → Grok 4.6. The same page says «No action is required to remove the models after they have been deprecated.», so nothing has to be cleaned up by hand; under the default model enablement, the alternatives the vendor names are enabled automatically for Copilot Enterprise and Copilot Business customers unless an administrator turned off the global default or explicitly disabled the model. A related date is written on OpenAI's Models page: «On October 14, 2026, GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on all plans», followed by the line saying it does not apply to the OpenAI API, and the model card on the same page says GPT-5.5 «remains available on the OpenAI API».

2026-10-22

The earliest shutdown date in Google's deprecations table for these preview models is 2026-10-22: veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview and gemini-omni-flash-preview, with gemini-omni-1.1-flash named as the replacement. The table carries a note above it: the shutdown dates it lists are the earliest possible dates, meaning nothing happens before that day rather than the models certainly stopping on it. Requests that call these previews directly should be moved before then. A related floor: the retirement floor written for claude-haiku-4-5-20251001 in Anthropic's deprecations table is 2026-10-15; that model is still Active (QCode does not currently support it), and the entry says the vendor will not consider retiring it sooner than that day, not that it stops on that day.

2026-10-23

OpenAI's shutdown day. Re-checked on 2026-10-10, the deprecations page carried 17 rows for the models removed from the API on 2026-10-23, 6 of them ft- fine-tuned entries, each with a replacement: gpt-4o-2024-05-13, the gpt-4 and gpt-4-turbo rows, o1 and o3-mini point to gpt-5.6-sol, gpt-4.1-nano points to gpt-5.6-luna, which is not currently supported on QCode, so use another GPT model such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra, o4-mini and gpt-3.5-turbo point to gpt-5.6-terra, and gpt-image-1 points to gpt-image-2.5-sunburst or gpt-image-2.5-flare. No gpt-6 model, and not gpt-6.1-sol, is on the list. The day before, 2026-10-22, is the earliest shutdown date in Google's table for veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview and gemini-omni-flash-preview (replacement gemini-omni-1.1-flash). After today there is still the mid-November batch, see the confirmed column. The newest notice on the deprecations page is the 2026-10-14 GPT-4 series deprecation, shutdown 2027-01-14, with gpt-6-luna in the replacement column; QCode does not currently support gpt-6-luna, so please use another GPT model.

Layers: published date / published hedge / floor / unverified

Dates officially announced and confirmed

Confirmed and quotable as of 2026-10-19: on 2026-10-13 Gemini 4 Argon (gemini-4-argon) became generally available (GA), and the vendor calls it «our most capable model built for complex coding, agentic, math, and science». On 2026-10-14 GPT-5.5 retired from ChatGPT, ChatGPT Work and Codex on all plans, with the official sentence «This retirement does not apply to the OpenAI API.» 2026-10-15 is the «Not sooner than October 15, 2026» floor for the Haiku 4.5 row (claude-haiku-4-5-20251001) in Anthropic's table, whose status still reads Active. On 2026-10-17 DeepSeek released DeepSeek-V4.1-Flash-1017; the release post is titled «Wish-Driven Programming Made Real», the vendor calls it «a Pareto improvement in performance and cost», and it is now live on the website, the app and the API; the changelog reads «DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017.» and «The API calling method and pricing remain unchanged — simply use deepseek-flash to access the latest model.». On 2026-10-19 GitHub Copilot drops Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. 2026-10-22 is the earliest shutdown date Google lists for three Veo 3.1 previews and gemini-omni-flash-preview. On 2026-10-23 OpenAI shuts down a batch of older models (list in the timeline). GitHub Universe takes place on 2026-10-28–29 at Fort Mason Center, San Francisco, in person and virtually. On 2026-11-16 GitHub Copilot drops Claude Opus 4.8 (still available to individual annual-plan Copilot subscribers), Claude Opus 5 and GPT-5.6 Terra, with the official alternatives Claude Opus 5.5, Claude Opus 5.5, and GPT-6.1 Sol or GPT-6 Luna; the announcement names every Copilot experience (Copilot Chat, inline edits, the ask and agent modes, and code completions) and schedules the models for 2026-11-16; individual annual-plan Copilot subscribers keep Claude Opus 4.8. Added at the 2026-10-13 re-check: the plans Ultrafast runs on — the Pricing page names its row «Ultrafast (Pro $500 and eligible Enterprise/Edu plans)», and the Models and Speed pages state the same eligibility — plus the official guide's sentence «Ultrafast mode for GPT-6 Astra and GPT-6.1 Sol is available to all API users.». Already happened: DevDay and the gpt-6.1-sol release on 2026-09-29; the four Copilot removals on 2026-10-02; the Mistral Large 4 public preview, the Decisions API beta and the usage-tier change on 2026-10-06; and the entry OpenAI's API changelog added on 2026-10-08, which reads «Added Ultrafast mode for GPT-6.1 Sol in the Responses API.». Still not confirmed: Mistral says the Large 4 weights come «by the end of the month», and the DevDay Exchanges cities carry no date or ticket price.

Unverified and unpublished (no rumours)

Where a vendor wrote no date, none is added. Ultrafast for GPT-6.1 Sol is no longer such a hedge: at the 2026-10-13 re-check the official Models and Speed pages both name the tier, and the quoted sentences, the plans and the date on the API side are listed in the confirmed column and in the OpenAI comparison card below rather than repeated here. Still unpublished: Mistral says the Large 4 weights come «by the end of the month»; collaborative slides keep the «in the coming weeks» wording from the 2026-09-29 launch, and no newer date was found; the DevDay Exchanges cities still carry no date or ticket price. The Decisions API, which read «in the coming days» on 2026-09-29, shipped in beta on 2026-10-06 and has moved to confirmed; Claude Haiku 5.5, which had nothing but two undated mentions as part of the 5.5 family (2026-09-22 and 2026-09-28), was released on 2026-10-07 under model ID claude-haiku-5-5 and is therefore no longer an unverified item here. Outlet reports and founder posts about the next DeepSeek or Grok release stay off the calendar.

The two vendors date things differently

Anthropic: retirement floors, no advance launch dates

The deprecations table gives a «Not sooner than …» floor per row (2026-10-15 for Haiku 4.5, 2027-06-30 for Sonnet 5, 2027-09-22 for Opus 5.5), while upcoming models got only an undated mention in the family. No sentence gave an exact release date then; Claude Haiku 5.5 has one now — released on 2026-10-07, model ID claude-haiku-5-5.

OpenAI: dates sit in the table

OpenAI's deprecations page gives concrete shutdown dates — the 2026-10-23 batch, the text-to-speech models on 2027-01-06, gpt-5.3-codex and others on 2027-04-01 — with the replacement model id in the next column. For new capability it starts with wording: the Decisions API, «in the coming days» on 2026-09-29, shipped in beta on 2026-10-06; Ultrafast for GPT-6.1 Sol began as an undated hedge, «Ultrafast support for GPT-6.1 Sol is coming later.» as the official Models page read on 2026-09-30. As of 2026-10-13 that hedge has become a row in a table and a day in a changelog: the Models page reads «Ultrafast is available», the Speed page reads «Ultrafast supports GPT-6 Astra and GPT-6.1 Sol.», the plans are Pro $500 and eligible Enterprise and Edu plans, and the date on the API side is 2026-10-08.

How to use this table

Three kinds of dates touch code: the 2026-10-23 shutdown list (requests that still name gpt-4o, gpt-4-turbo, o1, o3-mini, o4-mini or gpt-image-1 have to migrate), Anthropic's «not sooner than» floors (the Haiku 4.5 entry reads 2026-10-15), and requests that call Google's Veo 3.1 previews or gemini-omni-flash-preview directly (earliest shutdown 2026-10-22). GitHub Copilot needs a settings change instead: switch the models leaving on 2026-10-19 to the official alternatives; the four from 2026-10-02 are already gone. The 2026-10-14 GPT-5.5 entry is about ChatGPT, ChatGPT Work and Codex, and the official text states the OpenAI API is out of scope, so it asks for no code change. The 2026-11-16 batch (Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra) needs no manual cleanup, and the official text says the alternatives are enabled automatically for Copilot Enterprise and Copilot Business customers under the default model enablement. DeepSeek upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017 and the changelog states that the calling method and the pricing stay unchanged. The newest notice on the same table is the 2026-10-14 GPT-4 series deprecation, shutdown 2027-01-14, with gpt-6-luna in the replacement column; QCode does not currently support gpt-6-luna, so please use another GPT model, such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra. The target tier is the one the vendor writes in the right-hand column of its own table, and this schedule copies it as written rather than choosing one for you.

The QCode side: what these dates mean for calls here

The items on this calendar that touch QCode calls are these; for GPT calls on QCode, use the models listed on /models. The old ids on the 2026-10-23 shutdown list: the work belongs to requests whose code still names them, migrated along the official right-hand column; on the replacement side gpt-5.6-sol and gpt-5.6-terra are callable on QCode; QCode does not currently support gpt-6-luna or gpt-5.6-luna, so please use another GPT model, such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra; for the image models, /models is authoritative. claude-haiku-4-5-20251001: not currently supported on QCode, and 2026-10-15 is only the earliest day Anthropic will consider retiring it. Claude Opus 4.7: removed from Copilot on 2026-10-02, but still Active in Anthropic's table; claude-opus-4-7 is callable on QCode, and so is claude-opus-5-5, the model behind Copilot's suggested Claude Opus 5.5. The mid-November Copilot batch: GitHub Copilot drops another batch of models, and of the alternatives the vendor names, Claude Opus 5.5 corresponds to claude-opus-5-5 and GPT-6.1 Sol to gpt-6.1-sol, both callable on QCode; the other alternative is written GPT-6 Luna, which QCode does not currently support; please use another GPT model. Mistral Large 4 and the Gemini models are not offered on QCode. Billing is per token; see /models for each model's price. Want to use one of the models QCode offers today for a few days first? Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days (top-ups start at $5). If you already have a Personal Plan, buy a booster pack when the plan's quota runs short; Pay-as-you-go Packs are for accounts without an active plan.

Questions

What is certain to happen in October?

As of 2026-10-19, the dated October items are these: GPT-5.5 leaves ChatGPT, ChatGPT Work and Codex on 2026-10-14 (not the OpenAI API); 2026-10-15 is Haiku 4.5's «not sooner than» floor; GitHub Copilot drops this batch of models on 2026-10-19; 2026-10-22 is the earliest shutdown date for several Google preview models; OpenAI's older-model shutdown is on 2026-10-23; GitHub Universe is on 2026-10-28–29. The items dated 2026-10-01, 2026-10-02 and 2026-10-06 have already happened (see above). Another batch of Copilot models leaves in mid-November, on 2026-11-16. Among the official pages checked here, none gives a dated new-model release for the rest of October.

Will Haiku 4.5 stop on October 15?

No — Anthropic does not say that. The row reads «Not sooner than October 15, 2026», and at the re-check the status column still said Active; if that changes, the same table will carry the change. QCode does not currently support claude-haiku-4-5-20251001; the models callable on QCode are listed on /models.

When exactly did Haiku 5.5 come out?

Anthropic had named Haiku 5.5 as part of the 5.5 family twice (2026-09-22 and 2026-09-28), both times without a date; it released on 2026-10-07, model ID claude-haiku-5-5, with a 1M context window and 128K max output. Pricing has two rows by prompt length: prompts up to 100,000 tokens pay $0.10 input / $0.50 output per million tokens, longer prompts pay $0.50 input / $2.50 output for the whole request. claude-haiku-5-5 has been callable on QCode since 2026-10-09 — same key, same endpoint (https://api.qcode.cc/api), with claude-haiku-5-5 in the model field; QCode bills per token and per-model rates are listed on /models.

Does OpenAI's shutdown list affect my QCode requests?

The work that matters is in code still naming the old ids on the official list of models removed from the API on 2026-10-23: the official table gives a replacement on the right, and of those gpt-5.6-sol and gpt-5.6-terra are callable on QCode; gpt-5.6-luna, the listed replacement for ids such as gpt-4.1-nano, is not currently supported on QCode, and neither is gpt-6-luna, so those requests should use another GPT model, such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra; for the image models, /models is authoritative. How long the old ids themselves stay callable on QCode follows /models as well.

Should I sign up for DevDay Exchanges?

This page won't decide for you. About the Exchanges OpenAI published only city names — no schedule, no ticket price. The main event, by contrast, is fully specified: 2026-09-29, Fort Mason, $650 registration for invitees, keynote livestreamed free.

What about the next DeepSeek, Gemini or Grok?

That one is not listed. As of the 2026-10-19 check, none of the three published a release date for its next generation: at the 2026-09-30 check DeepSeek's own updates page did not point at a V5, and it now ends on V4.1-Flash-1017; Gemini 4 Argon went GA on 2026-10-13, and the earlier «much earlier than end of year» line was only outlet paraphrase; Grok 5 only in a founder's post. Leads, not dates.

Sources

Official pages fetched on 2026-10-19: DeepSeek's release post and changelog (DeepSeek-V4.1-Flash-1017 of 2026-10-17), the GitHub changelog post of 2026-10-15 (the 2026-11-16 deprecation notice) and Google's Gemini API release notes (Gemini 4 Argon GA of 2026-10-13) with its deprecations table (snapshot of 2026-10-17). Official pages fetched on 2026-10-10: OpenAI's API changelog (the two Oct 6 entries), its deprecations page (the 2026-10-01 notices and the 17 rows dated 2026-10-23), the Models and Pricing pages on learn.chatgpt.com (the GPT-5.5 retirement sentence) and devday.openai.com (Exchanges cities); Anthropic's model deprecations table and models overview; Mistral's Mistral Large 4 announcement and model docs; the GitHub changelog posts of 2026-09-18 and 2026-10-02 on Copilot model removals, plus githubuniverse.com. The 2026-09-29 gpt-6.1-sol launch post and recap are cited from Wayback snapshots of 2026-09-29 (direct access was blocked). Re-fetched on 2026-10-13 for the Ultrafast item alone: the Models, Speed and Pricing pages on learn.chatgpt.com, and the Ultrafast mode guide plus the API changelog on developers.openai.com (the entry dated 2026-10-08); the coming later line is quoted from a Wayback snapshot of that same Models page dated 2026-09-30； on the QCode side only the callable range is stated, and /models is authoritative.

The models already on the calendar are callable today

gpt-6-astra, gpt-6.1-sol, gpt-6-sol and claude-opus-5-5 share one QCode key (all callable on 2026-10-19, the check date); QCode does not currently support gpt-6-luna, so please use another GPT model.

Models and live rates View plans

gpt-5.3-codex, gpt-5.1, gpt-5.4-nano shut down 2027-04-01

Which model replaces each of the three, how to handle snapshots and default effort, and what changes in Codex (official table checked 2026-10-08)

Claude Haiku 5.5 status tracker

Released on 2026-10-07 as claude-haiku-5-5, with a 1M context window and two price rows; the tracker page keeps the official sentences and these fields item by item.

November 2026 model release calendar

Sonnet 4.5 retirement, GPT-5.6 Sol promo pricing, OpenAI shutdowns and three conferences, each with its source and check date.

Crypto payment guide

Nine coins, 56 networks, zero platform fee for AI coding API credits.

Checked on 2026-10-19 (each official table and page carries its own check date, see the sources; the Ultrafast plans and the 2026-10-08 API changelog line were re-checked on 2026-10-13). Only items a vendor actually wrote make it onto the schedule: a published date, a published time wording, or a published «not sooner than» / «earliest possible» floor. Items without a written date are left out, and quoted hedges stay in the vendor's own language so that «not sooner than» does not silently become «on that day» and «by the end of the month» does not become a specific day. The official pages are authoritative; model availability follows /models. QCode is independent of OpenAI, Anthropic, Mistral, GitHub and Google.