🔍 Read the full analysis: GPT-4o, GPT-4.1, GPT-5-nano: OpenAI API Retirement Dates And Migration Costs on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI’s deprecations page lists API shutdowns on Dec. 11, 2026, for GPT-5 models and several reasoning models, and Jan. 14, 2027, for GPT-4o and GPT-4.1 snapshots. The source reports large price differences between some retiring models and their listed replacements, alongside changes to reasoning behavior and model instructions. Teams should verify the official schedule and pricing, inventory model IDs, and test replacements before switching.
OpenAI’s API deprecation schedule sets Dec. 11, 2026, for the retirement of GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 Pro, o3 and o3 Pro, followed by Jan. 14, 2027, for listed GPT-4o and GPT-4.1 snapshots. Developers using those model IDs will need to change their API requests, and the replacement models may have different prices and response behavior.
The schedule, as reported from OpenAI’s deprecations page and checked on Nov. 1, 2026, shows that some models have already been shut down. On Oct. 23, OpenAI retired, among others, gpt-4.1-nano, the original GPT-4o snapshot, GPT-4 Turbo, GPT-3.5 Turbo, o1, o3 mini and o4 mini. Calls using those IDs may already fail. OpenAI says it emails affected customers and records deprecations on its public page.
For the December retirements, the listed replacements are from the GPT-5.6 family: GPT-5 and o3 move to GPT-5.6 Sol; GPT-5 mini to GPT-5.6 Terra; and GPT-5 nano to GPT-5.6 Luna. The January notice covers dated GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots, with GPT-6 Luna listed as the replacement. The source says fine-tuned versions follow their base model’s shutdown schedule, while the listed replacements do not support fine-tuning.
The source’s comparison of standard short-context prices per million tokens reports a sharp difference across some migrations. GPT-4o is listed at $2.50 input and $10 output, compared with $0.10 and $0.50 for GPT-6 Luna. GPT-5 nano is listed at $0.05 input and $0.40 output, while GPT-5.6 Luna is listed at $0.20 and $1.20. These are price figures checked Nov. 1, 2026, not a guarantee of future rates; the source says they apply to prompts up to 272,000 tokens.
Infographic 01 · The shutdown calendar
What stops working, and when
OpenAI's API deprecations, as listed on 1 November 2026. GPT-4o and GPT-4.1 already left ChatGPT on 13 February 2026; this round is about apps built on the API.
- 23 Oct 2026Already offAlready shut down
gpt-3.5-turbo,gpt-4,gpt-4-turbo,gpt-4.1-nano,gpt-4o-2024-05-13,o1,o1-pro,o3-mini,o4-mini,gpt-image-1. Calls to these fail today. - 30 Nov 2026Your testsEvals platform, Agent Builder, reusable prompts
Not models, but tools many teams test with. Existing evals became read-only on 31 Oct. Export them first.
- 11 Dec 2026Notified 11 JunGPT-5 family and o3
gpt-5→gpt-5.6-sol·gpt-5-mini→gpt-5.6-terra·gpt-5-nano→gpt-5.6-luna·o3→gpt-5.6-sol·gpt-5-pro,o3-pro→gpt-5.6-sol(pro mode) - 14 Jan 2027Notified 14 OctGPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini
Snapshots
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→gpt-6-luna. Fine-tuned versions follow their base model. - Jan to Apr 2027Plan aheadAudio, speech, older GPT-5.x
6 Jan:
tts-1and others · 20 Jan: older audio and realtime models · 26 Feb:whisper-1, GPT-4o transcription · 1 Apr:gpt-5.1,gpt-5.4-nano,gpt-5.3-codex
Sources: OpenAI deprecations page (checked 1 Nov 2026) · OpenAI Help Center, model release notes. Dates as listed by OpenAI; shutdowns can be rescheduled.
Powered by Thorsten Meyer AI
Migration Costs and App Behavior
The change is not just a matter of replacing a string in a configuration file. A move to a different model can affect API spending, output formats and application logic, so developers may need to revise budgets and tests as well as model IDs. The reported prices show that migration costs vary: some GPT-4-family users could pay less per token at the listed replacement rates, while GPT-5 nano users would face higher listed input and output rates.
The source also describes the replacements as reasoning models, which can change how applications use tokens. It says hidden reasoning tokens are billed as output, some sampling parameters must be removed, and a tight output limit may result in an empty answer that is still billed. Those are reported implementation concerns, not evidence that every application will encounter the same result. Teams should test their own prompts and workloads rather than assume the old and new models behave identically.
For businesses with multiple services, the deadlines create an inventory problem: model IDs can sit in code, tests, workflow tools or integrations maintained by different teams. The source also notes that OpenAI plans to shut down its Evals platform, Agent Builder and reusable prompt objects on Nov. 30, 2026. That date may affect teams preparing migration tests with those tools.
Products Worth Considering
Jackery Explorer 300 Portable Power Station, 292Wh
LG UltraGear 32GX850A-B 32-inch 4K UHD Glossy OLED Gaming Monitor
Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink
As an affiliate, we earn on qualifying purchases.
How the Retirement Calendar Builds
The schedule extends beyond the two headline dates. The source reports that OpenAI announced the Dec. 11 retirements on June 11 and the Jan. 14 retirements on Oct. 14. It says the October shutdown list included models announced for retirement in April. The source also notes that ChatGPT removals are separate: GPT-4o, GPT-4.1 and GPT-4.1 mini were removed from ChatGPT on Feb. 13, 2026, while this schedule concerns API access used by third-party applications.
OpenAI’s page, as described in the source, promises at least six months’ notice for generally available models, except where safety or compliance concerns require a faster timetable. The source observes that the interval between the October notice and the January retirement is three months, but says the notice does not explain which rule applies. OpenAI is also reported to say some customers may be able to buy dedicated capacity to retain access after a shutdown.
Other listed changes extend into 2027: text-to-speech models are scheduled for Jan. 6; older audio and realtime models for Jan. 20; Whisper and certain transcription models for Feb. 26; and GPT-5.1, GPT-5.4 nano and GPT-5.3 Codex for Apr. 1. These dates are based on the source’s Nov. 1 account of the deprecations page and should be checked against OpenAI’s current listing.
Notice Periods and Replacement Details
Several points remain unresolved in the source material. It does not say why the January retirements received three months’ notice despite the reported six-month policy for generally available models, or whether a safety or compliance exception applies. Customers with contracts or committed spending may need to ask OpenAI directly about how the schedule affects their arrangements.
Prices and model availability can change after the source’s Nov. 1, 2026, check. The reported cost comparison is limited to short-context standard rates; it does not calculate a customer’s total bill, which depends on token volume, caching, workload and any other applicable pricing terms. The source also does not provide independent tests showing how closely each replacement matches the retiring model for specific tasks.
The account reports changes to the Model Spec dated Oct. 29, including the removal of much of its detailed style guidance and the prior default to Markdown formatting. It does not establish how those changes affect every model or application. Developers should treat output changes as a risk to test, not as a confirmed outcome for every integration.
Audit IDs Before Each Deadline
Teams should first search application code, configuration files, test fixtures, automations and vendor integrations for every model ID in use. Check the current OpenAI deprecations page and confirm the relevant dates with the company, particularly if a contract, fine-tuned model or dedicated-capacity arrangement is involved. The source reports that some old IDs are already retired, so an inventory can also identify requests that may be failing now.
Before each shutdown, test the listed replacement against representative workloads. Compare answer quality, formatting, latency, token use, error handling and total cost; update parameters that the replacement does not accept and review output limits where reasoning tokens may count toward billing. If the application relies on a specific format, place that requirement explicitly in its instructions and verify it in tests. The immediate milestones are the reported Nov. 30 tooling shutdown, then the Dec. 11 and Jan. 14 model retirements.
Key Questions
When are GPT-5 nano and GPT-5 mini scheduled to leave the API?
The source reports a Dec. 11, 2026, retirement date for both. It lists GPT-5.6 Luna as the replacement for GPT-5 nano and GPT-5.6 Terra for GPT-5 mini. Check OpenAI’s current deprecations page before making changes.
When do the listed GPT-4o and GPT-4.1 API snapshots retire?
The reported date is Jan. 14, 2027, for the listed GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots. GPT-6 Luna is named as their replacement in the source.
Will migration necessarily make API calls more expensive?
No single cost outcome applies to every model. The source’s listed rates show GPT-6 Luna below GPT-4o’s standard short-context input and output rates, but GPT-5.6 Luna above GPT-5 nano’s. Actual bills depend on usage and current pricing.
Can fine-tuned versions keep running after the base model retires?
According to the source, OpenAI says fine-tuned models follow the shutdown schedule of their base model. It also reports that the listed replacement models do not support fine-tuning, so affected users should confirm their options with OpenAI.
What should developers test before switching?
Test representative prompts and application flows for output quality, required formatting, accepted parameters, token use, billing and error handling. Also search for retired model IDs across code, tests, automations and third-party integrations.
Source: ThorstenMeyerAI.com
