gpt-5.4-cyber has been removed from the OpenAI API: what you need to do now
The gpt-5.4-cyber model ID was removed from the OpenAI API on 1 October 2026. Calls to it now return errors. Migrate to gpt-5.6-cyber.
The deadline has passed
As of 1 October 2026, the gpt-5.4-cyber model ID no longer exists in the OpenAI API. OpenAI first flagged the deprecation on 11 September 2026, giving developers just under three weeks to migrate. That window has now closed. Any application still making calls to gpt-5.4-cyber will receive errors in response, not completions.
If you have not yet migrated, this is the moment to act. There is no grace period beyond the hard shutdown date.
What went and what replaces it
The documented replacement for gpt-5.4-cyber is gpt-5.6-cyber, described by OpenAI as the most capable cyber-tier model available to your account. The migration is a model ID swap in your configuration or code. There are no announced API surface changes between the two versions that would require deeper integration work, though you should validate outputs against your use case after switching.
OpenAI’s stated reason for the removal follows the same rationale given across its broader 2026 deprecation programme: improve reliability and simplify model selection by consolidating traffic onto newer, actively maintained versions.
What the three-week notice period tells you
For most generally available OpenAI models, the standard deprecation notice is at least six months. For specialised variants, including chat, Codex, and deep research tiers, it drops to at least three months. For preview models, it can be as short as two weeks.
The gpt-5.4-cyber removal sat at the shorter end of that range, with roughly 20 days between the deprecation announcement and the hard shutdown. If your team relies on cyber-tier models in production, that compressed timeline is worth factoring into your on-call and monitoring practices. A deprecation notice that arrives on a Friday afternoon can become an incident by the following month if no one is watching.
This sits inside a much larger 2026 deprecation wave
The gpt-5.4-cyber removal is one data point in a dense period of API changes. As of 30 September 2026, OpenAI had 117 upcoming dated retirements and 102 past retirements across its first-party API and partner routes. The second half of 2026 has been particularly active.
Key upcoming dates developers should be tracking:
| Shutdown date | What is removed | Recommended replacement |
|---|---|---|
| 23 October 2026 | gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4o-2024-05-13 | gpt-5.5 / gpt-5.4-mini |
| 23 October 2026 | o1, o1-pro, o3-mini, o4-mini | gpt-5.5 / gpt-5.5-pro |
| 31 October 2026 | Evals platform (read-only) | Full shutdown 30 November 2026 |
| 30 November 2026 | Agent Builder, Evals API, Prompts API | Agents SDK |
| 1 December 2026 | gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest | gpt-image-2 |
| 11 December 2026 | gpt-5-2025-08-07, o3-2025-04-16, o3-pro-2025-06-10 | gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna |
| 26 February 2027 | whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe | gpt-live-transcribe / gpt-transcribe |
The OpenAI deprecations page is the authoritative source and is actively updated. Third-party trackers such as benchlm.ai and benchr.org aggregate this into a cross-provider view if you are managing deprecations across multiple model providers.
What this means for your team
If your application calls gpt-5.4-cyber, update the model ID to gpt-5.6-cyber and deploy. Test your outputs before treating the migration as complete, because capability differences between model versions can affect downstream behaviour even when the API surface is identical.
Beyond this immediate fix, the broader deprecation calendar warrants a proper audit of every model ID your organisation uses in production, staging, and any automated pipelines. The 23 October 2026 batch removes some of the most widely used legacy IDs, including gpt-4 and gpt-3.5-turbo. If those strings appear anywhere in your infrastructure, you have roughly three weeks from 1 October 2026 to migrate them as well.
It is also worth noting that fine-tuning access has been tightened progressively through 2026. From 7 May 2026, new fine-tuning jobs became unavailable to organisations that had not previously run fine-tuning. From 2 July 2026, organisations needed to have run inference on a fine-tuned model within the previous 60 days to create new jobs. The ability to create new fine-tuning jobs for active existing customers is scheduled to end on 6 January 2027.
The pattern across all of this is consistent: OpenAI is consolidating its model catalogue onto the gpt-5.5 and gpt-5.6 families and retiring older identifiers on firm, published schedules. The practical implication is that model IDs should be treated as versioned dependencies in your codebase, with the same monitoring and update discipline you would apply to any other external dependency that can break without warning.
The OpenAI API changelog and the deprecations page are the two documents worth bookmarking and checking regularly if you build on these APIs.