Back to HomeAI API

OpenAI Model Shutdown Schedule (Late 2026 to 2027): When gpt-5, o3, GPT-4o, and gpt-image-2 Stop Working and What to Switch To

18 min min read
#OpenAI#OpenAI API#Model Deprecation#Model Migration#GPT-5#o3#GPT-4o#gpt-image-2#Whisper#GPT-6.1 Sol#GPT-6 Luna#Enterprise Procurement

Illustration of an OpenAI API shutdown timeline: older models on three lanes for text, images, and voice go dark one after another, and each point has an arrow to the newer model OpenAI recommends

Three Terms First: Deprecated, Shutdown, and Notice Period

OpenAI's official deprecations page groups models by status, and this schedule uses two of them throughout:

  • Deprecated: OpenAI has announced the model's retirement, but you can still call it until the shutdown date
  • Shutdown: the model no longer accepts requests through the standard API

The minimum notice periods on the official page (unless safety or compliance concerns require a faster timeline):

Model typeMinimum notice
Generally available (GA) modelsAt least 6 months
Specialized variants of GA models (e.g., chat, Codex, deep research) and image generation modelsAt least 3 months
Preview models with preview in the namePossibly much shorter; the page gives 2 weeks as an example

OpenAI emails customers who are actively using the model and documents the deprecation on the page. It also suggests calling the List models endpoint (GET /v1/models) and checking each model's optional shutdown_date field, which you can wire into your own monitoring. In some cases developers can contact sales to provision dedicated capacity for continued access after the shutdown date, but that isn't a standard option.

⚠️ Unlike Gemini API, where 2.5 Pro/Flash are automatically routed to newer models, OpenAI's deprecations page says nothing about automatic redirects. Once the shutdown date arrives, unchanged code simply fails.

Master Table: Shutdown Dates from December 2026 to April 2027

Compiled from the "Upcoming deprecations" section of the official page. The official table has 34 rows; here they're merged by shutdown date and replacement into 16 rows covering 9 shutdown dates. All dates are as listed on the official page, and none has passed as of today (2026-11-19).

ShutdownNoticeModels (including dated snapshots)Recommended replacement
2026-12-012026-06-02gpt-image-1-mini, gpt-image-1.5 (incl. -2025-12-16), chatgpt-image-latestgpt-image-2.6-sunburst or gpt-image-2.6-flare
2026-12-112026-06-11gpt-5 (incl. -2025-08-07), o3 (incl. -2025-04-16)gpt-6.1-sol
2026-12-112026-06-11gpt-5-mini (incl. -2025-08-07)gpt-6.1-terra
2026-12-112026-06-11gpt-5-nano (incl. -2025-08-07)gpt-6-luna
2026-12-112026-06-11gpt-5-pro (incl. -2025-10-06), o3-pro (incl. -2025-06-10)gpt-6.1-sol with reasoning.mode: pro in the Responses API
2027-01-062026-10-01tts-1, tts-1-hd, gpt-4o-mini-tts (incl. -2025-03-20, -2025-12-15)gpt-realtime-2.1-mini
2027-01-142026-10-14gpt-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-2025-04-14gpt-6-luna
2027-01-202026-07-20gpt-realtime (incl. -2025-08-28), gpt-4o-realtimegpt-realtime-2.1
2027-01-202026-07-20gpt-realtime-mini (incl. -2025-12-15), gpt-4o-mini-realtimegpt-realtime-2.1-mini
2027-01-202026-07-20gpt-audio (incl. -2025-08-28), gpt-audio-mini (incl. -2025-12-15), gpt-4o-audio, gpt-4o-mini-audiogpt-audio-1.5
2027-01-202026-07-20gpt-4o-mini-transcribe-2025-03-20gpt-4o-mini-transcribe-2025-12-15
2027-02-012026-11-05gpt-5.5-cyber-preview"The most capable cyber model available to you" (no model named)
2027-02-102026-11-10gpt-image-2 (incl. -2026-04-21)gpt-image-2.6-sunburst or gpt-image-2.6-flare
2027-02-262026-08-26whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe (incl. -2025-12-15), gpt-4o-transcribe-diarizegpt-live-transcribe or gpt-transcribe
2027-04-012026-10-01gpt-5.1 (incl. -2025-11-13), gpt-5.3-codexgpt-6-sol
2027-04-012026-10-01gpt-5.4-nano (incl. -2026-03-17)gpt-6-luna

Source: OpenAI's official deprecations page, checked 2026-11-19. Note that gpt-4o-mini-transcribe-2025-12-15, the recommended replacement in the 01-20 row, is itself on the 02-26 shutdown list, so it's only a stopgap.

Feature and Endpoint Deprecations

Beyond models, the page also lists several feature and endpoint changes:

DateNoticeItemWhat OpenAI says
2026-11-302026-06-03/v1/prompts and reusable prompt objectsDeprecated; move prompt content into your application code
2026-11-302026-06-03Evals dashboard, API, and gradersRead-only; OpenAI suggests migrating to Promptfoo
2026-11-302026-06-03Agent BuilderDeprecated; use the Agents SDK or ChatGPT Workspace Agents; ChatKit remains available
2027-01-062026-05-07Self-serve fine-tuning job creationRestricted access: existing customers can no longer create new fine-tuning jobs; inference on fine-tuned models stops only when the base model is deprecated
2027-01-112026-11-18GPT-5.2, GPT-5.4, and GPT-5.4 mini with EU data residency and service_tier: fast or priorityDeprecated; use the same models with EU data residency and Standard processing (service_tier: default), or the GPT-6.2 model family
2027-02-262026-08-26/v1/audio/translationsDeprecated; use /v1/realtime/translations

Batch by Batch: Where a Rename Isn't Enough

2026-12-11: The gpt-5 Family and o3 Shut Down on 2026-12-11

This is the batch with the most users. The official deprecations page currently recommends:

  • gpt-5, o3 → gpt-6.1-sol
  • gpt-5-mini → gpt-6.1-terra
  • gpt-5-nano → gpt-6-luna
  • gpt-5-pro, o3-pro → gpt-6.1-sol with reasoning.mode: pro set in the Responses API

The most common trap is tool calling. The official model pages for both gpt-6.1-sol and gpt-6.1-terra state that "Tool calling requires the Responses API," so code that uses Chat Completions with function calling needs to switch APIs, not just the model string. For gpt-6-luna, Chat Completions supports function calling only when reasoning effort is none.

gpt-6.1-sol supports reasoning.effort values low, medium (default), high, xhigh, and max; when a single request's input exceeds 272K tokens, input and cache prices double and output costs 1.5x. At the same price there is also GPT-6.2 Sol, released on 2026-11-17 ($2/$10, positioned by OpenAI as the go-to model for everyday work), worth evaluating if you'd rather jump to the newest generation in one move. See the GPT-6.2 Astra and Sol guide for details.

2027-01-14: GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini

All four are to be replaced with gpt-6-luna ($0.10/$0.50, positioned as "our most efficient model for focused, high-volume tasks"). Two things to watch:

  1. The list names dated snapshots: gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, and gpt-4.1-mini-2025-04-14. If your code uses an undated name, check shutdown_date with GET /v1/models rather than guessing
  2. Fine-tuned GPT-4o mini and GPT-4.1 shut down too: OpenAI states that fine-tuned versions follow the base model's schedule, and the gpt-6-luna model page marks Fine-tuning as not supported. Combined with the end of self-serve fine-tuning job creation on 2027-01-06, workflows built on fine-tuning need a replacement plan early

2027-01-06 and 01-20: Voice Models Change the Whole API

  • tts models: tts-1, tts-1-hd, and gpt-4o-mini-tts are to be replaced by gpt-realtime-2.1-mini, but its official page marks v1/audio/speech (speech generation) and Chat Completions as not supported; only v1/realtime is. A flow that sends text in one HTTP request and gets an audio file back has to be rewritten as a Realtime API session
  • Realtime and audio models: gpt-realtime-2.1 supports only v1/realtime; gpt-audio-1.5 supports only Chat Completions (Realtime, Responses, and v1/audio/speech are not supported). Move each old model to the replacement on the same endpoint

2027-02-26: whisper-1 and Transcription Models

The two replacements support different endpoints, and picking the wrong one fails outright:

  • gpt-transcribe: positioned as a "high-accuracy speech-to-text model for file and Realtime input transcription"; supports v1/audio/transcriptions and Realtime transcription sessions, at $0.0045 per minute. Use this one for uploaded audio files
  • gpt-live-transcribe: positioned as a "low-latency speech-to-text model for realtime transcription"; v1/audio/transcriptions is marked not supported, so it only works through Realtime transcription sessions, at $0.017 per minute

On the same day /v1/audio/translations (speech-to-English translation) is deprecated; the official replacement is /v1/realtime/translations.

2026-12-01 and 2027-02-10: Image Models

gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest shut down first on 12-01, followed by gpt-image-2 on 2027-02-10. The replacement for all of them is gpt-image-2.6-sunburst or gpt-image-2.6-flare. Both official model pages list the same prices (per million tokens: text input $5, image input $8, image output $30), and Flare is positioned for "image generation and editing for fast iteration." Both support only the Images API for generation (v1/images/generations) and editing (v1/images/edits); v1/responses is marked not supported.

2027-04-01: gpt-5.1, gpt-5.3-codex, gpt-5.4-nano

gpt-5.1 and gpt-5.3-codex are to be replaced with gpt-6-sol, and gpt-5.4-nano with gpt-6-luna. Note that the gpt-6-sol model page itself says "See GPT-6.1 Sol for the newer Sol model." gpt-6-sol, gpt-6.1-sol, and GPT-6.2 Sol all cost $2/$10 for input/output; the difference is cached input ($0.20 for gpt-6-sol, $0.10 for the other two). Since you have to change the code anyway, it's worth evaluating the newer versions directly.

Replacement Specs and Pricing

ModelOpenAI positioningInput / cached input / cache write / output (per million tokens)Context / max input / max outputKnowledge cutoff
gpt-6.1-solNear-Astra performance at a lower cost$2/$0.10/$2.50/$101,050,000 / 922,000 / 128,0002026-04-30
gpt-6.1-terraVisual reasoning, precise computer use, natural writing$1/$0.05/$1.25/$51,050,000 / 922,000 / 128,0002026-06-01
gpt-6-lunaMost efficient, for focused, high-volume tasks$0.10/$0.01/$0.125/$0.501,050,000 / 922,000 / 128,0002026-05-18
gpt-6-solComplex coding and agentic workflows$2/$0.20/$2.50/$101,050,000 / 922,000 / 128,0002026-04-20
(For reference) GPT-6.2 SolGo-to model for everyday work$2/$0.10/$2.50/$101,050,000 / 922,000 / 128,0002026-06-01
ModelUsePriceSupported endpoints
gpt-image-2.6-sunburst / gpt-image-2.6-flareImage generation and editingPer million tokens: text input $5 (cached $1.25), image input $8 (cached $2), image output $30Images API generation and editing
gpt-realtime-2.1-miniFaster, lower-cost realtime voicePer million tokens: text $0.6/$2.4, audio $10/$20v1/realtime only
gpt-realtime-2.1Reasoning voice model with tool usePer million tokens: text $4/$24, audio $32/$64v1/realtime only
gpt-audio-1.5Audio in, audio out on Chat CompletionsPer million tokens: text $2.5/$10, audio $32/$64Chat Completions only
gpt-transcribeHigh-accuracy transcription$0.0045 per minutev1/audio/transcriptions, Realtime transcription
gpt-live-transcribeLow-latency realtime transcription$0.017 per minuteRealtime transcription only

Sources: OpenAI's official model page for each model (checked 2026-11-19); the GPT-6.2 Sol row follows the GPT-6.2 Sol official model page. USD, Standard billing.

Migration Cost Comparison (Arithmetic, Not Benchmarks)

This section compares only the options after you switch; it doesn't estimate what the old models cost. All scenarios assume Standard billing, no cache writes, no extra tokens from reasoning, and single-request inputs under 272K.

Scenario A: gpt-5 or o3 Text Work, 50M Input and 10M Output per Month

OptionInput costOutput costMonthly total
gpt-6.1-sol (official recommendation)50M × $2 = $10010M × $10 = $100$200
GPT-6.2 Sol50M × $2 = $10010M × $10 = $100$200
gpt-6.1-terra50M × $1 = $5010M × $5 = $50$100
gpt-6-luna50M × $0.10 = $510M × $0.50 = $5$10

If 30M of the 50M input tokens are cache hits: gpt-6.1-sol comes to 20M × $2 + 30M × $0.10 + $100 = $143; gpt-6.1-terra comes to $20 + $1.50 + $50 = $71.50. Whether a cheaper model is good enough only shows up when you run your own workload; this is price only.

Scenario B: gpt-5.1 or gpt-5.3-codex, 100M Input per Month (80M Cache Hits) and 10M Output

OptionUncached inputCached inputOutputMonthly total
gpt-6-sol (official recommendation)20M × $2 = $4080M × $0.20 = $1610M × $10 = $100$156
gpt-6.1-sol$4080M × $0.10 = $8$100$148
GPT-6.2 Sol$4080M × $0.10 = $8$100$148

The heavier your caching, the wider the gap between gpt-6-sol and the two newer models, since cached input costs twice as much.

Scenario C: High-Volume Classification on GPT-4o mini, 200M Input and 20M Output per Month

On gpt-6-luna: 200M × $0.10 = $20, 20M × $0.50 = $10, $30 per month.

Scenario D: whisper-1 Transcription, 10,000 Minutes per Month

OptionMathMonthly total
gpt-transcribe10,000 × $0.0045$45
gpt-live-transcribe10,000 × $0.017$170

The realtime model costs about 3.8 times as much as the file model. For batch transcription of audio files there's no reason to choose the realtime one, and it doesn't support v1/audio/transcriptions anyway.

Migration Checklist

  1. Inventory first: search your code and config for gpt-5, o3, gpt-4o, gpt-4.1, tts-1, whisper-1, gpt-image, gpt-realtime, gpt-audio, gpt-5.1, gpt-5.3-codex, and gpt-5.4-nano, and check each model's shutdown_date with GET /v1/models
  2. Prioritize against the master table: the 12-01 image models and the 12-11 gpt-5/o3 batch are most urgent; leave time for regression testing
  3. Confirm the replacement supports your endpoint: Chat Completions with tool calling → move to the Responses API for gpt-6.1-sol/gpt-6.1-terra; v1/audio/speech → Realtime API; v1/audio/transcriptions → gpt-transcribe; images go through the Images API
  4. Change parameters for the pro models: move gpt-5-pro and o3-pro to gpt-6.1-sol with reasoning.mode: pro in the Responses API
  5. Reset reasoning effort: gpt-6.1-sol defaults to medium with options from low to max; gpt-6-luna needs none for function calling on Chat Completions
  6. Plan fine-tuned models separately: fine-tuned versions shut down with their base model; self-serve fine-tuning job creation ends on 2027-01-06, and gpt-6-luna doesn't support fine-tuning
  7. Rebudget long prompts: on gpt-6.1-sol and GPT-6.2, single-request input over 272K tokens doubles input and cache prices and makes output 1.5x; the gpt-6-sol model page also says prompts over 272K cost more
  8. EU data residency users: if you run GPT-5.2/5.4/5.4 mini with fast or priority tiers, switch to service_tier: default or the GPT-6.2 family before 2027-01-11
  9. Features that end on 11-30: if you still use /v1/prompts, Evals, or Agent Builder, those will be deprecated or go read-only on 2026-11-30
  10. Wire shutdown dates into monitoring: OpenAI emails active users, but the email may land in an unwatched account; checking shutdown_date regularly is safer

CloudInsight Enterprise OpenAI API Procurement | Local Invoices, No Foreign Credit Card Needed

This round of retirements touches several models at once, and the replacements are priced differently. What enterprises get stuck on most is often not the technology but payment, invoicing, and budget control.

Contact CloudInsight for an OpenAI API Enterprise Quote

We offer one-stop procurement for Claude, GPT, and Gemini, Taiwan uniform invoices, and technical support in Chinese.

How to Choose: By the Model You Use Today

Using nowOfficial recommendationCheck before switching
gpt-5, o3gpt-6.1-solTool calling needs the Responses API; GPT-6.2 Sol at the same price is also worth evaluating
gpt-5-minigpt-6.1-terraTool calling needs the Responses API
gpt-5-nano, gpt-5.4-nano, the GPT-4o family, the GPT-4.1 familygpt-6-lunaNo fine-tuning; effort must be none for function calling on Chat Completions
gpt-5-pro, o3-progpt-6.1-sol + reasoning.mode: proMust use the Responses API
gpt-5.1, gpt-5.3-codexgpt-6-solgpt-6.1-sol and GPT-6.2 Sol cost the same with half-price cached input
tts-1, tts-1-hd, gpt-4o-mini-ttsgpt-realtime-2.1-miniNo v1/audio/speech; move to the Realtime API
whisper-1, the gpt-4o-transcribe familygpt-transcribe or gpt-live-transcribeChoose gpt-transcribe for uploaded audio files
gpt-image-1.5, gpt-image-2, etc.gpt-image-2.6-sunburst or -flareImages API only

FAQ: OpenAI Model Shutdowns

When do gpt-5 and o3 shut down?

gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, and o3-pro (including dated snapshots) will shut down on 2026-12-11. The notice date was 2026-06-11.

What should replace gpt-5?

The official deprecations page currently recommends gpt-6.1-sol for gpt-5 and o3, gpt-6.1-terra for gpt-5-mini, gpt-6-luna for gpt-5-nano, and gpt-6.1-sol with reasoning.mode: pro in the Responses API for gpt-5-pro and o3-pro.

When will GPT-4o stop working?

gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, and gpt-4.1-mini-2025-04-14 will shut down on 2027-01-14, and OpenAI recommends gpt-6-luna.

Will requests be redirected to the new model after shutdown?

The official deprecations page doesn't say so. It defines shutdown as the model "no longer accepts requests through the standard API"; the replacements are recommendations, and you have to change your code yourself.

When does gpt-image-2 shut down?

gpt-image-2 (including gpt-image-2-2026-04-21) will shut down on 2027-02-10, with notice given on 2026-11-10; the replacement is gpt-image-2.6-sunburst or gpt-image-2.6-flare. The older gpt-image-1.5, gpt-image-1-mini, and chatgpt-image-latest will shut down on 2026-12-01.

How do I know whether my model has a shutdown date?

Call the List models endpoint (GET /v1/models) and check each model's optional shutdown_date field. OpenAI also emails customers who are actively using the model.

Conclusion: Handle the Two December Batches First; Voice and Fine-Tuning Take the Most Work

Three Core Recommendations

  1. Finish migrating off gpt-5/o3 before December 11: they shut down on 2026-12-11, and code that uses tool calling also has to move from Chat Completions to the Responses API
  2. Schedule voice, transcription, and fine-tuning work early: the tts replacement only runs on the Realtime API, gpt-live-transcribe doesn't accept file uploads, and gpt-6-luna doesn't support fine-tuning; none of these is a one-string change
  3. If you're changing anyway, pick the newest generation: gpt-6-sol, gpt-6.1-sol, and GPT-6.2 Sol share the same input/output price, and the newer two charge half as much for cached input

Ready to Plan Your Migration?

Contact the CloudInsight sales team and we'll help inventory the models you need to replace, plan your usage mix on the replacements, and handle payment and invoicing.

Join our LINE Official Account for instant model-selection consultation.


Further Reading

References

Need Professional Cloud Advice?

Whether you're evaluating cloud platforms, optimizing existing architecture, or looking for cost-saving solutions, we can help

Book Free Consultation

Related Articles

AI API

GPT-6 Astra, 6.1 Sol, Luna: API Pricing, Specs, and Which to Pick

OpenAI shipped GPT-6 Astra, GPT-6 Sol and Luna, and GPT-6.1 Sol within one month in September 2026. This guide covers the official API pricing for all four models ($10/$50, $2/$10, $0.10/$0.50), the long-prompt pricing rule behind the 1M context window, real cost math, and the code settings to change when migrating from GPT-5.6 — now also covering GPT-6.1 Terra ($1/$5), added on October 26.

AI API

Gemini 2.5 Pro and Flash Auto-Route From December 1: Which Model You Get, Pricing, and the Code to Change (2026)

On 2026-11-02 Google announced that starting 2026-12-01, every API request to gemini-2.5-pro will automatically route to gemini-3.8-flash and every request to gemini-2.5-flash to gemini-3.5-flash-lite, billed at the new models' rates. This is not a shutdown, and requests won't fail because of the model name, but function calling validation, thinking settings, sampling parameters, and prefilled turns all have breaking changes. This guide covers the routing map, prices before and after, cost math, and the code to change before go-live.

AI API

GPT-6.2 Astra and Sol Guide: API Pricing, Specs, and Which to Pick (2026)

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 2026-11-17. Astra costs $8/$40 per million tokens, 20% less than GPT-6 Astra's $10/$50, with cached input down from $1 to $0.40; Sol is $2/$10, the same as GPT-6.1 Sol. This guide covers official specs, the surcharge for inputs over 272K tokens, Batch/Fast/Ultrafast multipliers, cost math that includes the long-prompt trap, and the settings to change when you switch.