Gemini 2.5 Pro, Flash, and Flash-Lite retire on 20 October 2026 on Vertex AI: here is what to do before then

If you have production workloads running on Google’s Gemini Enterprise Agent Platform (formerly Vertex AI) and you are still calling gemini-2.5-pro, gemini-2.5-flash, or gemini-2.5-flash-lite, you have a hard deadline to plan for: 20 October 2026. After that date, any API call to those model IDs returns a 404 error. No graceful degradation, no automatic routing to a newer model. Just a broken integration.

Here is everything you need to know to handle the migration cleanly.

The correct date is 20 October, not 16 October

You may have seen 16 October 2026 cited in earlier communications, including emails Google sent to affected customers. That date appeared on Google’s Gemini API deprecations page between roughly May and August 2026, and it referred to the Gemini Developer API. It has since been removed from that page entirely.

The official model versions and lifecycle page for the Gemini Enterprise Agent Platform lists 20 October 2026 as the retirement date for all three models. That is the authoritative figure for anyone using Google Cloud.

To be clear about scope: this retirement applies to the Gemini Enterprise Agent Platform API (formerly the Vertex AI Gemini API). The Gemini Developer API, which you access with an API key rather than a Google Cloud project, lists no hard shutdown date for 2.5 models at the time of writing, though access has been restricted to projects that actively used them before late July 2026.

What happens on 20 October 2026

Google defines a “retired” model as one that is no longer accessible or supported. From that date, calling any of the three model IDs from a Google Cloud project will return:

{
  "error": {
    "code": 404,
    "message": "This model models/gemini-2.5-pro is no longer available to new users. Please update your code to use a newer model for the latest features and improvements."
  }
}

Worth noting: new Google Cloud projects have already been receiving this error for months. If your project has not called a 2.5 model in a while, you may find you are already blocked well ahead of the official date. The 90-day inactivity rule means that projects flagged as inactive lose access sooner.

Which models to migrate to

Google’s lifecycle page maps each retiring model to specific replacements:

Retiring model	Retirement date	Recommended replacements
`gemini-2.5-pro`	20 Oct 2026	`gemini-3.8-flash`, `gemini-3.5-flash`
`gemini-2.5-flash`	20 Oct 2026	`gemini-3.8-flash`, `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite`
`gemini-2.5-flash-lite`	20 Oct 2026	`gemini-3.8-flash`, `gemini-3.1-flash-lite`, Gemma 4

All of the Gemini 3 replacements are generally available on the platform now. As a reference point, gemini-3.8-flash was released on 2 September 2026 with a retirement floor of no earlier than 2 September 2027, giving you at least a year of runway from the point you migrate.

A few things worth flagging as you choose:

There is no direct Pro successor yet. Google does not list a Gemini 3 Pro model as a GA replacement for gemini-2.5-pro. Gemini 3.1 Pro is still in preview. The official recommendation points gemini-2.5-pro users toward gemini-3.8-flash or gemini-3.5-flash, which are capable Flash-tier models rather than Pro-tier equivalents.

Costs will change. If you are running gemini-2.5-flash-lite workloads at scale, be aware that moving to the cheapest like-for-like option still represents a meaningful price increase. Run your token volumes through Google’s pricing page before committing to a replacement.

gemini-3.5-flash has its own timeline to watch. This model was deprecated on the Gemini API on 8 October, with requests there now routed to gemini-3.6-flash. On Vertex, gemini-3.6-flash itself retires on 19 November. If you pick gemini-3.5-flash as your migration target, check that timeline and plan accordingly.

What to do if you have Provisioned Throughput

If your workloads run under a Provisioned Throughput (PT) agreement on Vertex AI, there is an extra step. Your PT assignment is pinned to a specific model endpoint. When that endpoint retires, your throughput allocation does not automatically transfer to the new model.

You need to submit a request to update your PT assignment to point at one of the replacement model endpoints. Google recommends doing this roughly a month before the retirement date, so by mid-September 2026, to allow time for the request to be reviewed and approved. If you leave this until the week before 20 October, you risk a gap in coverage.

The Provisioned Throughput supported models page lists which models currently accept PT orders.

The staggered shutdown picture

20 October 2026 is when retirement begins for the general population, but Google has published a more detailed staggered timeline for the full hard shutdown:

28 January 2027: gemini-2.5-flash-lite fully shut down
31 March 2027: gemini-2.5-flash and gemini-2.5-pro shut down across most regions
6 August 2027: Remaining regional deployments (KR, BR, FR data residency zones)

This means projects that have been actively using these models may have more runway than the headline date suggests. However, treating 20 October as your migration deadline is the safer approach. Projects with lower usage could be blocked at or before that point, and the PT migration process alone warrants getting ahead of it.

The SDK note

If you are still using the Vertex AI SDK, Google has indicated that SDK releases from June 2026 onwards do not support Gemini models. New Gemini features are only available through the Gen AI SDK. If you have not already moved to the Gen AI SDK, a model migration is a natural point to handle that switch at the same time.

What to do now

Audit your codebase for any hardcoded references to gemini-2.5-pro, gemini-2.5-flash, or gemini-2.5-flash-lite on Google Cloud.
Test your chosen replacement model against your actual workloads. Output quality and cost can vary meaningfully across model tiers.
If you have Provisioned Throughput, raise a migration request with Google by mid-September 2026 at the latest.
Update your SDK if you are still on the Vertex AI SDK.
Set a reminder for 20 October 2026 as a hard go-live deadline, even if the full shutdown is staggered beyond that date.

The models themselves are not disappearing overnight for everyone, but 20 October is the point at which Google offers no guarantees of continued access. Getting your migration done before then is the only way to avoid an unexpected outage.