Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- The cutoff is October 24, 2026 at 00:00 UTC, or 08:00 in China on the same day. The announcement says the affected models remain available until that cutoff.
- The list contains three GPT-5.4 IDs plus GPT-5.2, GPT-5.1, GPT-5 and GPT-5-mini. Updating only the primary model can leave retired IDs in a fallback chain.
- After the cutoff, retired IDs will no longer be accepted or returned by model-list endpoints. The announcement specifies neither an exact error code nor automatic remapping.
- The current Agent API catalog lists available OpenAI choices, but it provides no one-to-one replacement mapping for the seven retiring IDs.
Confirm the API scope and cutoff
This change concerns teams accessing models through Perplexity. The current official changelog names Agent API and Orchestrator API as the affected services, with retirement set for October 24, 2026 at 00:00 UTC. That is 08:00 in China on the same day; teams elsewhere should convert from the same UTC instant.
The changelog groups the notice under September 2026 without a specific announcement day. Schedule changes against the October 24 UTC cutoff; the announcement says the affected models remain available until then.
Check all seven IDs in selections and fallback chains
The three GPT-5.4 IDs are openai/gpt-5.4, openai/gpt-5.4-mini and openai/gpt-5.4-nano.
The other four are openai/gpt-5.2, openai/gpt-5.1, openai/gpt-5 and openai/gpt-5-mini. Match the complete provider-prefixed ID and suffix when reviewing configuration, so a search for a display name does not miss an entry.
Perplexity asks developers to update direct selections and fallback chains before the cutoff. The Agent API documentation uses model for a single selection and models for a fallback list. If a model fails or is unavailable, the API tries the next model in the chain, so rarely used backup entries also need an inventory.
Current choices are not a replacement mapping
As of October 14, 2026, the official Agent API catalog lists openai/gpt-6.1-sol, openai/gpt-6-sol and openai/gpt-6-luna among its OpenAI choices.
The same catalog also lists openai/gpt-5.6-sol, openai/gpt-5.6-terra, openai/gpt-5.6-luna and openai/gpt-5.5. These are current catalog options; the list does not designate any one of them as the official counterpart to a particular retiring ID.
Choose around the reasoning, tools and output budget your task needs. The documentation cautions that model capabilities vary. For Orchestrator API, also check the target against that API's current model catalog rather than assuming every Agent API option is available there.
Four configuration checks before migration
First, inventory affected Perplexity integrations. Search for the complete IDs in code, environment configuration, task templates and saved model overrides, recording an owner and each place they are used.
Second, inspect direct model selections and every models fallback list for all seven retiring IDs. Even where the primary model has changed, check whether backup entries still reference an old model.
Third, choose a current candidate for each use and review its capabilities, parameter requirements, output budget, service tier and Perplexity pricing. Do not substitute another provider's list prices for Perplexity's charges.
Fourth, schedule the configuration changes before the cutoff and retain the previous configuration, target IDs and an owner for subsequent acceptance checks. This checklist plans a migration; it does not establish that an arbitrary replacement has passed workload compatibility checks.
Hypothetical example: a changed primary can leave an old fallback
Suppose an integration has a task selecting openai/gpt-5.4-mini, while a separate fallback list still contains openai/gpt-5. Both IDs are retiring. Replacing only the task's direct selection leaves another configuration entry to migrate. This example illustrates the review scope rather than reporting an actual API result.
Choose currently supported candidates for that use and review the entire fallback list again. The presence of other models in a list does not establish uninterrupted service after the cutoff. The announcement does not promise to rewrite a retired ID automatically to a new model.
What changes at the cutoff, and what remains unspecified
Perplexity explicitly says retired IDs will no longer be accepted or returned by model-list endpoints after the cutoff. It does not specify the HTTP status or error text that will be returned. Retain the actual response when investigating a failure rather than treating one assumed error code as the only signal.
This is a Perplexity API lifecycle announcement. It does not establish the same retirement date for OpenAI's direct API or ChatGPT. The current catalog also does not guarantee unchanged quality, latency or cost after migration; those outcomes need evaluation for the intended workload.
Sources
Frequently Asked Questions
Does an old ID missing from the current catalog mean it has already retired?
That conclusion does not follow. The changelog says the affected models remain available until October 24, 2026 at 00:00 UTC. The catalog's displayed scope does not replace the specific retirement notice or establish that a particular account's current request will succeed.
Must integrations using presets manually replace all seven IDs?
First establish whether the integration uses a preset or an explicit model override. The retirement notice asks developers to update direct selections and fallback chains. It does not establish that every preset contains an old ID or that every preset user needs the same change.
Can a similar model name determine the replacement?
A similar name does not establish capability compatibility. Perplexity provides no one-to-one migration table. Choose a current model around tools, reasoning, output budget and task requirements, leaving time for subsequent acceptance checks.
What additional parameter is required for an Anthropic fallback?
The current Agent API documentation requires requests using an anthropic/* model to include max_output_tokens. Omitting it returns HTTP 400. This required condition applies specifically to Anthropic models and should not be generalized to every model.
Does an unsupported service tier in a fallback list cause rejection?
The current documentation says flex or priority applies only when the selected model, or every model in the fallback list, supports it. Otherwise, the API ignores service_tier and uses default processing. The response's service_tier reports the tier actually served, so the request configuration alone does not establish processing or cost.