OpenAI legacy model retirement: the 09-28 batch has passed
next is 2026-10-23
As of 2026-10-20, the 2026-09-28 shutdown date on OpenAI's deprecations table has passed for gpt-3.5-turbo-instruct, babbage-002, davinci-002 and gpt-3.5-turbo-1106, all with gpt-5.6-terra as replacement. The next batch is 2026-10-23: gpt-3.5-turbo → gpt-5.6-terra, gpt-4 and gpt-4-turbo → gpt-5.6-sol, gpt-4.1-nano → GPT-5.6 Luna and gpt-image-1 → GPT Image 2.5; fine-tuned versions such as ft-babbage-002 and ft-davinci-002 go the same day. Finish config changes before 2026-10-23.
Updated 2026-10-20
One key across Claude, GPT and Chinese models, billed per token.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard / AMEX), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four separate facts
Shutdown date of the Completions-era ids
Official table: 2026-09-28 for gpt-3.5-turbo-instruct, babbage-002, davinci-002 and gpt-3.5-turbo-1106, all → gpt-5.6-terra. As of 2026-10-07 this section sits under the Past deprecations heading.
Next shutdown date
The 2026-04-22 section: gpt-3.5-turbo (snapshot gpt-3.5-turbo-0125) → gpt-5.6-terra; gpt-4 (snapshot gpt-4-0613) and gpt-4-turbo → gpt-5.6-sol; gpt-4.1-nano → GPT-5.6 Luna; gpt-image-1 → GPT Image 2.5 Sunburst or Flare; o1-pro → gpt-5.6-sol (reasoning.mode: pro). The table writes the date as October 23, 2026.
The batch after that
Notified on 2026-06-11: o3-2025-04-16 and gpt-5-2025-08-07 → gpt-5.6-sol; gpt-5-mini-2025-08-07 → gpt-5.6-terra; gpt-5-nano-2025-08-07 → GPT-5.6 Luna; gpt-5-pro-2025-10-06 and o3-pro-2025-06-10 → gpt-5.6-sol (reasoning.mode: pro), shutdown 2026-12-11.
Already scheduled for 2027
Announced 2026-10-01: gpt-5.1 and gpt-5.3-codex → gpt-6-sol, gpt-5.4-nano → gpt-6-luna, with six months' notice according to OpenAI. A further batch on 2027-01-14: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini → gpt-6-luna.
What to do once a shutdown date has passed
As of 2026-10-20 the table's 2026-09-28 shutdown date for those old ids has passed, and by OpenAI's own definition a model or endpoint is no longer accessible at the time of the shut down. If you still call gpt-3.5-turbo-instruct, babbage-002, davinci-002 or gpt-3.5-turbo-1106, switch model to gpt-5.6-terra now. Then check the next batch: gpt-3.5-turbo, gpt-4 and gpt-4-turbo shut down on 2026-10-23, gpt-4.1-nano moves to GPT-5.6 Luna and gpt-image-1 to GPT Image 2.5; each row names its own replacement, so do not map them all to one id.
What the official table says as of 2026-10-20
In the deprecations page re-fetched on 2026-10-20, the four 2026-09-28 rows of the 2025-09-26 section (Legacy GPT model snapshots) now sit under Past deprecations, all replaced by gpt-5.6-terra. The 2026-04-22 section now appears above Past deprecations, its text still reads “Access to these models will be shut down on the dates below”, and its rows are dated October 23, 2026: gpt-3.5-turbo-0125 with aliases gpt-3.5-turbo and gpt-3.5-turbo-completions → gpt-5.6-terra; gpt-4-0613 with aliases gpt-4, gpt-4-0613-completions and gpt-4-completions → gpt-5.6-sol; gpt-4-turbo → gpt-5.6-sol; gpt-4.1-nano → GPT-5.6 Luna; gpt-image-1 → GPT Image 2.5 Sunburst or Flare. Its fine-tuning table maps ft-gpt-3.5-turbo, ft-babbage-002 and ft-davinci-002 to gpt-5.6-terra, ft-gpt-4 to gpt-5.6-sol and ft-gpt-4.1-nano-2025-04-14 to GPT-5.6 Luna.
Timeline
Date in the official section title (2025-09-26: Legacy GPT model snapshots). The text says OpenAI is deprecating older models with declining usage over the next six to twelve months.
Re-fetched the deprecations page on 2026-10-20: the four 2026-09-28 rows remain under Past deprecations, replacement still gpt-5.6-terra; the 2026-04-22 section now sits above Past deprecations in the upcoming list, still dated October 23, 2026.
Next shutdown date: gpt-3.5-turbo → gpt-5.6-terra, gpt-4 and gpt-4-turbo → gpt-5.6-sol, and fine-tuned versions such as ft-babbage-002 and ft-davinci-002 are removed the same day.
Confirmed vs caution
Officially announced / confirmed
Readable word for word on the deprecations page fetched 2026-10-20: 2026-09-28 shuts gpt-3.5-turbo-instruct, babbage-002, davinci-002 and gpt-3.5-turbo-1106, replacement gpt-5.6-terra; October 23, 2026 shuts the gpt-3.5-turbo, gpt-4 and gpt-4-turbo snapshots and aliases, replacements gpt-5.6-terra or gpt-5.6-sol; gpt-4.1-nano → GPT-5.6 Luna, gpt-image-1 → GPT Image 2.5 Sunburst or Flare; at the time of the shut down the model or endpoint is no longer accessible.
Caution / not published
First, as of 2026-10-20 the 2026-04-22 section sits above Past deprecations on the official page, consistent with its October 23, 2026 dates; it is no longer placed under Past deprecations as it was on 2026-10-07. Second, when QCode starts rejecting a retired id is not published; plan to finish before 2026-10-23. Third, OpenAI says that in some cases developers may provision dedicated capacity for continued access after the shutdown date; that is a case-by-case option, not a postponement. Fourth, the table maps ids only; whether outputs match is yours to test.
The 09-28 wave versus the 10-23 wave
2026-09-28: Completions-era ids (date passed)
gpt-3.5-turbo-instruct, babbage-002, davinci-002, gpt-3.5-turbo-1106 → gpt-5.6-terra. The table date has passed; if you still call them, swap the id first.
2026-10-23: next batch (gpt-3.5-turbo / gpt-4 / gpt-4-turbo)
gpt-3.5-turbo → gpt-5.6-terra, gpt-4 and gpt-4-turbo → gpt-5.6-sol, gpt-4.1-nano → GPT-5.6 Luna, gpt-image-1 → GPT Image 2.5. o1, o3-mini and o4-mini share the same date and have their own page.
Three steps before 2026-10-23
① Search: grep code, environment variables and scripts for gpt-3.5-turbo-instruct, babbage-002, davinci-002, gpt-3.5-turbo-1106 and the next batch gpt-3.5-turbo, gpt-4, gpt-4-turbo, aliases together with snapshots (gpt-4-0613 and gpt-4 share one cell). ② Swap: change model per the row, gpt-5.6-terra for terra rows and gpt-5.6-sol for sol rows; ft-babbage-002 and ft-davinci-002 map to gpt-5.6-terra. ③ Verify: send one minimal request and confirm the model in the response is the new id.
What to do on QCode
As of 2026-10-20, the official replacements gpt-5.6-terra and gpt-5.6-sol are callable on QCode, as is gpt-6-sol from the 2027-04-01 batch; 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. GPT models use the OpenAI endpoints and cannot use the Anthropic endpoint: the OpenAI SDK base_url is https://api.qcode.cc/openai/v1 (requests go to /openai/v1/chat/completions), and the Codex CLI base_url is https://api.qcode.cc/openai (requests go to /openai/v1/responses). After swapping the old id, send one non-streaming request and check the model in the reply. Billing is per token; per-model prices are 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.
FAQ
Has the 2026-09-28 batch already shut down?
The table's 2026-09-28 date has passed (as of 2026-10-20), and the section is listed under Past deprecations. By OpenAI's definition, at the time of the shut down the model or endpoint is no longer accessible; what the API actually returns is up to OpenAI.
When is the next batch and what is in it?
2026-10-23. The 2026-04-22 section lists snapshots and aliases including gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4o-2024-05-13, gpt-4.1-nano, o1, o3-mini, o4-mini and gpt-image-1, dated October 23, 2026. Aliases such as gpt-3.5-turbo-completions and gpt-4-completions share their snapshot's cell, date and replacement.
What replaces gpt-3.5-turbo and gpt-4?
gpt-3.5-turbo (snapshot gpt-3.5-turbo-0125) → gpt-5.6-terra; gpt-4 (snapshot gpt-4-0613) → gpt-5.6-sol; gpt-4-turbo → gpt-5.6-sol. Follow the row and do not remap everything to one id.
Are fine-tuned babbage-002 and davinci-002 models affected?
Yes. The fine-tuning table dates ft-babbage-002 and ft-davinci-002 October 23, 2026, replacement gpt-5.6-terra; ft-gpt-3.5-turbo → gpt-5.6-terra and ft-gpt-4 → gpt-5.6-sol. OpenAI also states that inference on fine-tuned models continues until the base models are deprecated.
What happens on 2026-12-11 and 2027-04-01?
2026-12-11: older GPT-5 and o3 snapshots notified on 2026-06-11, for example o3-2025-04-16 → gpt-5.6-sol, gpt-5-mini-2025-08-07 → gpt-5.6-terra and gpt-5-nano-2025-08-07 → GPT-5.6 Luna. 2027-04-01: announced 2026-10-01, gpt-5.1 and gpt-5.3-codex → gpt-6-sol, gpt-5.4-nano → gpt-6-luna. A further batch on 2027-01-14: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini → gpt-6-luna.
Do aliases like gpt-3.5-turbo-completions need separate handling?
There is no separate date. OpenAI writes gpt-3.5-turbo-completions and gpt-4-completions in the same cell as their snapshots, with the same date, October 23, 2026, and the same replacement. How to move the call itself is defined by OpenAI's documentation.
Sources
Shutdown dates and replacements: OpenAI's deprecations page https://developers.openai.com/api/docs/deprecations (fetched 2026-10-20; sections 2025-09-26, 2026-04-22, 2026-06-11, 2026-10-01 and 2027-01-14 plus the deprecation definitions). QCode setup: the docs.qcode.cc page on endpoints and API paths (fetched 2026-10-07, re-checked 2026-10-20). OpenAI writes dates as October 23, 2026, Dec 11, 2026 or Apr 1, 2027; this page renders them as ISO.
Swap old ids to the table's replacements before 2026-10-23
One key for gpt-5.6-terra, gpt-5.6-sol and gpt-6-sol; billed per token, per-model prices on /models.
Related
gpt-4-turbo retires 2026-10-23
The gpt-4-turbo snapshot and aliases shut the same day, replacement gpt-5.6-sol.
GPT-4 series deprecation and removal
Noticed 2026-10-14, removed 2027-01-14: four GPT-4 models, the snapshot IDs, the official replacement gpt-6-luna, plus shutdown_date and the migration steps.
Codex Spark retirement
A different retirement topic, for comparison.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
Checked 2026-10-20. Shutdown dates and replacements follow OpenAI's official deprecations page; model availability follows /models; changes may happen without notice. Not affiliated with OpenAI.