Is Vertex AI Deprecated? Imagen and Veo Endpoint Shutdown List for June 30, 2026

If you are asking "is Vertex AI deprecated?", the answer is narrower but urgent: Vertex AI itself is not being retired, but Google is shutting down a long list of Imagen and Veo generative-media endpoints on June 30, 2026. Every production Imagen endpoint — from imagegeneration@002 through imagen-4.0-ultra-generate-001 — and every Veo 2.x/3.0 GA video endpoint will stop accepting requests. If your routing config, SDK calls, or fallback chains reference any of these model IDs, they will break.

Vertex AI deprecation list: what is actually being retired

The scope is broader than most teams expect. Google is not sunsetting the Vertex AI platform — it is consolidating an entire generation of standalone image and video generation endpoints into the Gemini model family. If someone asks for a list of what is deprecated, these are the Vertex AI model IDs to remove from production configs:

14 Imagen endpoints retiring: imagegeneration@002, imagegeneration@003, imagegeneration@004, imagegeneration@005, imagegeneration@006, imagetext@001, imagen-3.0-capability-001, imagen-3.0-capability-002, imagen-3.0-fast-generate-001, imagen-3.0-generate-001, imagen-3.0-generate-002, imagen-4.0-fast-generate-001, imagen-4.0-generate-001, imagen-4.0-ultra-generate-001.

3 Veo endpoints retiring: veo-3.0-generate-001, veo-3.0-fast-generate-001, veo-2.0-generate-001.

The replacement path is also explicit: image generation requests move to gemini-2.5-flash-image. Video requests move to veo-3.1-generate-001 (or veo-3.1-fast-generate-001 for the fast tier).

Preview endpoints already retired on March 3, 2026 — if you are still running a -preview or -exp model ID, it was already shut down in April.

Why this deadline matters beyond "update your endpoint name"

The surface change looks simple: swap one model string for another. The underlying implications are more significant.

Capability model shift. Imagen was a purpose-built diffusion model for image generation. gemini-2.5-flash-image is a multimodal Gemini model with image generation as one output mode. The two behave differently on fine-grained text rendering, photorealism, and prompt adherence for complex compositions. If your application relies on Imagen-4.0-ultra for high-fidelity outputs, the Flash-tier Gemini model may not match it on every benchmark. Google's own migration guidance does not include a 1:1 quality parity promise.

Pricing structure change. Imagen endpoints were priced per image or per second of video. Gemini 2.5 Flash Image uses the standard Gemini token-based billing model, where output images consume tokens differently than text. If you built cost controls around Imagen's per-image pricing, your billing model changes on July 1.

Latency and throughput profile. Flash-class models are optimized for latency, not the same throughput ceilings as dedicated Imagen infrastructure. If you have burst workloads — batch avatar generation, product catalog automation, ad creative pipelines — the new endpoint may throttle differently.

Fallback chain design. Many teams set up Imagen-4.0 as their primary with Imagen-3.0-fast as a cost-optimized fallback. Both endpoints now collapse into a single gemini-2.5-flash-image target. Your fallback model no longer exists as a separate endpoint. This changes how you design degradation paths.

What is also deprecated but not yet urgent

Google also announced that Vertex AI Extensions is deprecated and will shut down after November 26, 2026, with migration to Agent Platform as the replacement. If your agent toolchain depends on Extensions for grounding, code execution, or data store access, you have more runway but should start evaluating the Agent Platform APIs now.

Gemini 2.5 Pro, Flash-Lite, and Flash retirement dates were also updated to October 16, 2026 — meaning the very model you are migrating to has its own eventual end date. Teams should build endpoint aliasing and version-pinned migration plans rather than hardcoding gemini-2.5-flash-image as a permanent target.

What routing teams should do this week

1. Audit every Vertex AI model ID in your codebase and config files. Search for imagegeneration@, imagen-3.0, imagen-4.0, veo-2.0, veo-3.0. Any match needs a migration ticket opened today.

2. Test gemini-2.5-flash-image against your actual prompts. Do not assume equivalence. Run your production prompt portfolio through the new endpoint and compare outputs before June 30, not after.

3. Update billing forecasts. Token-based pricing for image generation will produce different cost curves than per-image pricing. Model the delta before it shows up in your July invoice.

4. Rebuild fallback paths. With Imagen endpoints collapsing into a single Gemini target, your degradation strategy changes. Consider whether other providers (Replicate, Stability AI, Midjourney API) should serve as fallback tiers for image generation rather than relying on cross-model fallback within Vertex.

5. Set deprecation monitoring. Gemini 2.5 Flash Image itself retires October 16, 2026. Build your endpoint management so the next migration is a config change, not a code change.

The routing case

This is the third major endpoint consolidation at a hyperscaler in 2026, after AWS sunsetting older Bedrock model versions in Q1 and Azure deprecating specific GPT-4 preview endpoints in April. The pattern is consistent: providers consolidate older endpoint families into newer model-series endpoints, changing pricing, latency, and behavior profiles in the process.

For teams using a routing gateway, the migration is a config update to the provider mapping. For teams hardcoding Vertex model IDs directly into application code or CI/CD pipelines, it is a code change, a test cycle, and a deployment window — all with a 18-day deadline.

If you are evaluating routing infrastructure, this deprecation is a useful stress test for your architecture: can you swap a model endpoint in one place, or does it require touching every service that calls it?