OpenAI's October 2026 deprecation wave: GPT-4o and GPT-4.1 on 2027-01-14, TTS to gpt-realtime-2.1-mini

For anyone tracking OpenAI's API lifecycle, 2026-10-17 started with a precondition worth stating: developers.openai.com became reachable for this site again - the official deprecations page (api/docs/deprecations) fetched in full that day (a 504 KB page), where this site's previous successful read of the official domain dates to 2026-09-01. That made four announcements posted between 2026-09-11 and 2026-10-14 visible at once: three are future shutdown plans, one is already in effect. Every item below can be pointed to in the page text fetched on 2026-10-17.

Overview: four announcements, one table

The page adds one important line to the 2027-01-14 batch: Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models.

2027-01-14: the GPT-4o and GPT-4.1 families' last day

The section's framing sentence reads: On October 14, 2026, we notified developers using certain GPT-4 series models - a three-month span from notice to shutdown. All four family rows list the same replacement: gpt-6-luna.

Two boundaries are easy to get wrong:

gpt-4o-2024-05-13 is not in this batch. The earliest snapshot belongs to the older legacy GPT snapshot batch (announced 2026-04-22), shutting down on 2026-10-23 with tier-based GPT-5.6 replacements. Check the snapshot date before the batch - the two tables must not be merged.

The SDK annotations carry more aliases than the page. openai-python v3.32.0 (2026-10-16) marks the undated aliases gpt-4o, gpt-4o-mini, gpt-4.1, and gpt-4.1-mini in its 2027-01-14 row as well; the node v7.35.0 annotation lists only the five dated snapshots. The page's display-name-plus-snapshot style is compatible with both - treat the page as the source of truth and the annotations as the cross-check.

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

The page states the notice period explicitly: this batch was announced on 2026-10-01 and removed on 2027-04-01 with six months' notice. Replacements map by tier - gpt-5.3-codex and gpt-5.1 to gpt-6-sol, gpt-5.4-nano to gpt-6-luna.

The timing gap is worth recording: the SDKs' ChatModel deprecation annotations only gained the 2027-04-01 row in v3.32.0 / v7.35.0 on 2026-10-16, and they list gpt-5.4-nano and gpt-5.1 but not gpt-5.3-codex. The page announcement sat there for half a month before the SDK types caught up, and the SDK list is a strict subset of the page's - governance scripts keyed only to SDK types would miss the codex variant row.

2027-01-06: the TTS four-pack moves to Realtime

The same 2026-10-01 announcement batch retires text-to-speech as a line: tts-1, tts-1-hd, gpt-4o-mini-tts-2025-03-20, and gpt-4o-mini-tts-2025-12-15 shut down on 2027-01-06 with at least three months' notice. The single official replacement is gpt-realtime-2.1-mini, with the page pointing developers to the Realtime API guide to plan the migration.

This is the first time the text-to-speech direction gets an official destination - and it lands on a Realtime voice model, not a successor on the speech endpoint. For callers this is an endpoint-plus-model migration: gpt-realtime-2.1-mini speaks over the Realtime channel (WebRTC / WebSocket), so changing a model string is not the whole job. It sits adjacent to, but separate from, the later 2027-01-20 window for the legacy audio/realtime family (gpt-realtime, gpt-audio, and others).

gpt-5.4-cyber: the row already in effect

The only effective item in this wave: gpt-5.4-cyber was announced deprecated on 2026-09-11 and removed from the API on 2026-10-01 - more than two weeks past when this site checked (2026-10-17). The official replacement advice is a single sentence: The most capable cyber model available to you, with no specific model ID. Combined with the Daybreak tiering announced on 2026-08-07 (Blue for general defensive work, Red for separately approved specialized models), the available cyber surface depends on your account's approvals.

Migration checklist: three steps

1. Inventory: bucket every model string in production code and infrastructure by snapshot date - the 2026-10-23 batch (legacy snapshots, including gpt-4o-2024-05-13, gpt-4.1-nano, o1, o3-mini and others), the 2027-01-06 batch (the TTS four-pack), the 2027-01-14 batch (the newer GPT-4o / GPT-4.1 snapshots), and the 2027-04-01 batch (gpt-5.1 / gpt-5.3-codex / gpt-5.4-nano). SDK annotations help machine scanning, but the official page rules.
2. Map: follow the official replacements - GPT-4o / GPT-4.1 lines to gpt-6-luna; gpt-5.1 and gpt-5.3-codex to gpt-6-sol; gpt-5.4-nano to gpt-6-luna; TTS to gpt-realtime-2.1-mini (channel change included). Fine-tuned versions follow their base models - include the fine-tuning job list.
3. Schedule: the nearest deadline is 2027-01-06 (TTS, about two and a half months out), then 2027-01-14 (one week later). The notice-period rules and calendar maintenance live in the OpenAI Deprecation Lifecycle Guide: Notice Periods, Shutdown Calendar, Migration Checklist.

Next steps

• Continuously maintained timelines: OpenAI Models Release Notes (2026, Living Document) and OpenAI 2026 H2 Model and API Release Timeline.
• Background on the replacement tiers (the GPT-6 family landing in the SDKs): The GPT-6 family grows: gpt-6-sol, gpt-6-luna, gpt-6.1-sol land in the official SDKs.
• The channel fundamentals behind the TTS migration target: Realtime Voice guide: gpt-realtime and Voice Mode.
• The SDK-side carrier of this wave's annotations: openai-python 3.32.0 and openai-node 7.35.0: prompt-created voices, Live audio fixtures, and Decisions in Live.