Deprecation date table · as of 2026-11-20

AI API deprecations December 2026 to February 2027: the shutdown calendar

As of November 20, 2026, the three nearest dates to act on are November 30, 2026, when Anthropic removes fast mode for Claude Opus 4.8 in Claude Code, claude-sonnet-4-5-20250929 retires the same day with claude-sonnet-5-5 as the replacement, and OpenAI shuts down /v1/prompts and turns the Evals dashboard read-only; December 1, 2026, when gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest shut down; and December 11, 2026, when the gpt-5 and o3 families shut down. On the Claude side, three retirement dates fall inside the next three months: claude-opus-4-5-20251101 (Not sooner than November 24, 2026), claude-opus-4-6 (February 5, 2027) and claude-sonnet-4-6 (February 17, 2027). All three are still Active, because those are lower bounds, not retirement days.

Updated 2026-11-20

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

#Deprecation calendar#OpenAI model shutdowns#Claude model retirements#Not sooner than

Four dates to hold on to

11-30

Opus 4.8 fast mode removed

Anthropic's wording: Fast mode for Opus 4.8 was deprecated on October 30, 2026, and will be removed on November 30, 2026. Keep Opus 4.8 with fast mode on past that date and the API rejects your requests. The same day, OpenAI shuts down /v1/prompts and reusable prompt objects and turns the Evals dashboard and API read-only.

12-01

Three image models shut down

gpt-image-1-mini, gpt-image-1.5 (including the gpt-image-1.5-2025-12-16 snapshot) and chatgpt-image-latest shut down on December 1, 2026, with gpt-image-2.6-sunburst or gpt-image-2.6-flare as the recommended replacement; all three rows carry a Jun 2, 2026 notice date.

01-14

gpt-4o and gpt-4.1 shut down

On January 14, 2027 OpenAI shuts down GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20), GPT-4o mini (gpt-4o-mini-2024-07-18), GPT-4.1 (gpt-4.1-2025-04-14) and GPT-4.1 mini (gpt-4.1-mini-2025-04-14); every row names gpt-6-luna as the replacement and carries an Oct 14, 2026 notice date.

Not sooner than

A floor, not a retirement day

The Claude deprecation table is headed API model name, Current state, Deprecated and Tentative retirement date; an Active row carries N/A under Deprecated and a Not sooner than date under Tentative retirement date, for example claude-opus-4-5-20251101 with N/A in the Deprecated column and Not sooner than November 24, 2026 in the Tentative retirement date column, still Active.

Why it matters: hard-coded model IDs start failing on the shutdown date

As of November 20, 2026, Anthropic and OpenAI both publish the next three months of deprecations, shutdowns and removals, but one table is organised by model and the other by date, and the replacement names keep changing. Any project that hard-codes an old model ID in code, an environment variable or config.toml starts failing on the shutdown date: OpenAI defines Shutdown as the point where the model no longer accepts requests through the standard API, and Anthropic says requests to retired models fail.

This calendar merges both into one ordered list. The OpenAI model shutdown dates 2027 batch runs from January 6, 2027 to February 26, 2027; the Claude side shows lower bounds only.

New or imminent items between September 30, 2026 and November 20, 2026

Four items from this stretch are worth doing first.

  1. 1October 30, 2026: fast mode for Opus 4.8 enters deprecation and is removed on November 30, 2026; the official replacement is Opus 5.5, which needs Claude Code v2.1.280 or later, or you turn fast mode off in the session.
  2. 2October 14, 2026: OpenAI marks the GPT-4o and GPT-4.1 families Deprecated with a January 14, 2027 shutdown date and names gpt-6-luna in every row.
  3. 3November 5, 2026 and November 10, 2026: gpt-5.5-cyber-preview is set to shut down on February 1, 2027, and gpt-image-2 with gpt-image-2-2026-04-21 on February 10, 2027, still pointing at gpt-image-2.6-sunburst or gpt-image-2.6-flare.
  4. 4September 30, 2026: Anthropic notified developers that claude-sonnet-4-5-20250929 retires on November 30, 2026, replaced by claude-sonnet-5-5.

The three nearest dates

2026-11-30

Three things land on this one day.

  1. 1Anthropic removes fast mode for Opus 4.8 in Claude Code: to keep fast mode, switch with /model to Opus 5.5, which needs Claude Code v2.1.280 or later and leaves fast mode on; to stay on Opus 4.8, run /fast off for standard speed, otherwise requests carrying fast mode are rejected by the API.
  2. 2claude-sonnet-4-5-20250929 retires, replaced by claude-sonnet-5-5 (that one has its own page already, so it gets a single line here).
  3. 3OpenAI shuts down /v1/prompts and reusable prompt objects and turns the Evals dashboard and API read-only, recommending that prompt content move into application code and that evals move to Promptfoo.
2026-12-11

Two batches shut down on this date: gpt-5, gpt-5-2025-08-07 and o3, o3-2025-04-16 are replaced by gpt-6.1-sol; gpt-5-mini and gpt-5-mini-2025-08-07 by gpt-6.1-terra; gpt-5-nano and gpt-5-nano-2025-08-07 by gpt-6-luna; gpt-5-pro and o3-pro move to gpt-6.1-sol with reasoning.mode: pro in the Responses API. This batch has its own page already, so only the dates and replacements are recorded here.

2027-01-14

The GPT-4o and GPT-4.1 families shut down: GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20), GPT-4o mini (gpt-4o-mini-2024-07-18), GPT-4.1 (gpt-4.1-2025-04-14) and GPT-4.1 mini (gpt-4.1-mini-2025-04-14), all replaced by gpt-6-luna, with a notice date of Oct 14, 2026. Still ahead after that: the realtime and audio batch on January 20, 2027, gpt-5.5-cyber-preview on February 1, 2027, gpt-image-2 on February 10, 2027, and whisper-1 with the transcription family on February 26, 2027; each row is listed below.

Verified dates vs common misreadings

Verified: the dates on both official deprecation pages

Checked row by row on November 20, 2026. OpenAI: November 30, 2026 shuts down /v1/prompts and reusable prompt objects and turns the Evals dashboard, API and graders read-only; December 1, 2026 shuts down gpt-image-1-mini, gpt-image-1.5 (including gpt-image-1.5-2025-12-16) and chatgpt-image-latest, replaced by gpt-image-2.6-sunburst or gpt-image-2.6-flare; December 11, 2026 shuts down six rows, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro; January 6, 2027 shuts down tts-1, tts-1-hd and gpt-4o-mini-tts with its two snapshots, replaced by gpt-realtime-2.1-mini; January 14, 2027 shuts down the four GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini rows, replaced by gpt-6-luna; January 20, 2027 shuts down gpt-realtime, gpt-audio, gpt-4o-realtime, gpt-4o-audio, gpt-realtime-mini, gpt-audio-mini, gpt-4o-mini-realtime, gpt-4o-mini-audio and gpt-4o-mini-transcribe-2025-03-20, replaced by gpt-realtime-2.1, gpt-audio-1.5, the matching mini tiers and gpt-4o-mini-transcribe-2025-12-15; February 1, 2027 shuts down gpt-5.5-cyber-preview, whose replacement the page words as the most capable cyber model available to you, with no fixed ID; February 10, 2027 shuts down gpt-image-2 and gpt-image-2-2026-04-21; February 26, 2027 shuts down whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize, replaced by gpt-live-transcribe or gpt-transcribe.

Anthropic: claude-sonnet-4-5-20250929 retires on November 30, 2026; the floors are November 24, 2026 for claude-opus-4-5-20251101, February 5, 2027 for claude-opus-4-6 and February 17, 2027 for claude-sonnet-4-6, all three still Active; the October 15, 2026 floor for claude-haiku-4-5-20251001 has already passed and the model is still Active.

Common misreadings and what is not published

None of the following is an official conclusion.

  1. 1Treating the date after Not sooner than as a retirement day: it is the earliest floor, not a shutdown time, and the state stays Active; Anthropic only commits to at least 60 days' notice for publicly released models, so the real date arrives with a later notice.
  2. 2Updating a config from an old replacement name seen on a summary site: in the official table the gpt-5 family is replaced by gpt-6.1-sol, gpt-6.1-terra and gpt-6-luna, while some summaries still list gpt-5.6-* as the gpt-5 replacement, so follow the official row.
  3. 3How much notice OpenAI gives: the page only says affected customers are notified by email and in the documentation, with no number of days.
  4. 4Whether these dates slip: nothing is promised, the column is called Shutdown date, and changes appear on the page itself.

Waiting for the error vs switching ahead of the table

Change it on shutdown day: requests fail

OpenAI's Shutdown state means the model no longer accepts requests through the standard API; Anthropic's Retired state means requests to the model fail, and Claude Code on the Claude API shows an error that begins There's an issue with the selected model. A move to a newer model is also the moment to check temperature, top_p and top_k in the request: all three were removed in API version 2026-10-21 and sending them returns a 400 error, while earlier API versions return an error when they are set to a non-default value on Claude 4.7 and later models. Anthropic's advice is to omit the parameters and use prompting to guide model behavior.

Switch to the official replacement first, then run the regression

Work down the date table: map every old ID still in use to the replacement in the official row, change it on a branch, run the regression suite, then merge. Replacement models such as gpt-6-luna, gpt-image-2.6-sunburst and gpt-realtime-2.1 are not on the QCode list, so read them as vendor-published replacements; the models QCode carries, claude-opus-5-5, claude-sonnet-5-5, gpt-6.2-sol and gpt-6.1-sol, switch over with the same key.

Four steps: find the old IDs, swap them, point the endpoint, run the regression

  1. 1Search every place old IDs appear: repositories, environment variables, ~/.codex/config.toml and CI config, looking for strings such as gpt-5, o3, gpt-4o, gpt-4.1, tts-1, whisper-1 and claude-sonnet-4-5-20250929; on the Claude side, confirm which model actually runs even when no model is set or a Claude Code alias such as sonnet is in use.
  2. 2Replace per the official row: the four GPT-4o and GPT-4.1 families go to gpt-6-luna, gpt-5 and o3 go to gpt-6.1-sol, the three image models go to gpt-image-2.6-sunburst or gpt-image-2.6-flare; for Opus 4.8 fast mode, either switch with /model to Opus 5.5 (Claude Code v2.1.280 or later) or run /fast off.
  3. 3Run the same code through QCode: for Claude models set ANTHROPIC_BASE_URL to https://api.qcode.cc/api and ANTHROPIC_MODEL to claude-opus-5-5 or claude-sonnet-5-5; for Codex CLI put base_url = "https://api.qcode.cc/openai" and wire_api = "responses" in ~/.codex/config.toml with model set to gpt-6.2-sol.
  4. 4Run the regression suite, then re-read both official deprecation pages a week before each shutdown date, because dates and replacements are edited in place.

How to connect through QCode: one key for Claude and GPT

QCode is an independent API access service with no affiliation to Anthropic or OpenAI.

  1. 1Claude models use the Anthropic Messages endpoint only: set ANTHROPIC_BASE_URL to https://api.qcode.cc/api and switch with /model claude-opus-5-5 or /model claude-sonnet-5-5 in Claude Code; Opus 5.5, the official replacement for Opus 4.8 fast mode, maps to claude-opus-5-5 on QCode.
  2. 2Codex CLI uses the responses protocol: in ~/.codex/config.toml set base_url = "https://api.qcode.cc/openai" and wire_api = "responses", with model set to gpt-6.2-sol or gpt-6.1-sol.
  3. 3OpenAI-compatible clients such as the OpenAI SDK and Cline use base_url https://api.qcode.cc/openai/v1. Billing is per token, with live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.

Common questions

So which day does the Not sooner than date actually shut down?

Not necessarily the shutdown day: the page gives a floor. The Claude table is headed API model name, Current state, Deprecated and Tentative retirement date, and an Active row puts N/A under Deprecated and a Not sooner than date under Tentative retirement date: claude-opus-4-5-20251101 says Not sooner than November 24, 2026, claude-opus-4-6 says February 5, 2027 and claude-sonnet-4-6 says February 17, 2027, all three still Active. The real retirement date arrives with the later notice, and Anthropic commits to at least 60 days' notice for publicly released models.

Can I still use the model after that floor date?

Yes, as long as the state is still Active. claude-haiku-4-5-20251001 carries a floor of Not sooner than October 15, 2026, that date has already passed and the state is still Active; no row in the table turned Retired because of it. What actually fails is Retired: the model is no longer available for use and requests to retired models fail.

Does fast mode degrade gracefully when Opus 4.8 fast mode is removed?

No, requests are simply rejected. The docs say that after removal, requests still carrying fast mode on Opus 4.8 are rejected by the API rather than falling back to standard speed. Two ways out: switch with /model to Opus 5.5, which needs Claude Code v2.1.280 or later and keeps fast mode on, or run /fast off to stay on Opus 4.8 at standard speed.

What should gpt-4o be replaced with?

All four rows name gpt-6-luna: GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20), GPT-4o mini (gpt-4o-mini-2024-07-18), GPT-4.1 (gpt-4.1-2025-04-14) and GPT-4.1 mini (gpt-4.1-mini-2025-04-14), all shutting down on January 14, 2027. gpt-6-luna is the vendor-published replacement and is not on the QCode list; to run a GPT-family model on QCode, switch to gpt-6.2-sol, gpt-6.1-sol or gpt-6.1-terra, billed per token with rates on /models.

My config uses an alias, or no model at all. Does this still hit me?

Yes. Anthropic asks you to check which model you are actually using even if you set no model or rely on a Claude Code alias such as sonnet, then set its recommended replacement and change the old model everywhere a model is set, including subagent definitions. The version without the date suffix is hit too: the Sonnet 4.5 retirement notice says it includes claude-sonnet-4-5, the same model's ID without a date.

How do I hear about these shutdowns early?

There are only two official channels; the rest is self-audit. Anthropic notifies customers with active deployments, with at least 60 days' notice for publicly released models; OpenAI says affected customers are notified by email and in the documentation, without a number of days. The self-audit is in Anthropic's own steps: open the Usage page in Claude Console, click Export, and read the downloaded CSV to see usage broken down by API key and model.

Sources

Anthropic model deprecations, platform.claude.com/docs/en/about-claude/model-deprecations (checked 2026-11-20); Claude Code fast mode documentation, code.claude.com/docs/en/fast-mode (checked 2026-11-20); OpenAI API deprecations, developers.openai.com/api/docs/deprecations (checked 2026-11-20). Dates, model IDs and replacements are taken word for word from these three pages; no summary site was used.

Swap the old IDs for the official replacements, then run the same code on QCode

Create a QCode account for a cr_ key, set ANTHROPIC_BASE_URL=https://api.qcode.cc/api in Claude Code and base_url = "https://api.qcode.cc/openai" in Codex CLI, billed per token, with per-model rates on /models.

Related reading

Crypto payment guide

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

Dates and model IDs were checked on 2026-11-20 against the official pages at code.claude.com, platform.claude.com and developers.openai.com; vendors update these tables in place, so re-check before changing a config. QCode.cc is an independent API access service with no affiliation to and no endorsement from Anthropic or OpenAI; vendor deprecations and shutdowns follow each vendor's own announcement. QCode model availability and live rates follow qcode.cc/models.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack