Google's Gemini API deprecation schedule lists today — August 17, 2026 — as the shutdown date for all three Imagen 4 GA endpoints: imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001. The recommended replacement on the developer API is gemini-3.1-flash-image, Google's native image-generation model inside the Gemini 3.x line (branded "Nano Banana" in migration docs).
The same model IDs on Vertex AI carry a different calendar: Google's Vertex model page lists a June 30, 2026 discontinuation date for those Imagen 4 GA versions. Third-party lifecycle trackers that scrape only one surface can therefore show June 30 while the Gemini API page shows August 17 — same ID, different platform documentation.
What's new
- Gemini API shutdown (primary): Per Google's Gemini API deprecations page, all three Imagen 4 GA IDs released June 24, 2025 shut down August 17, 2026, replaced by gemini-3.1-flash-image.
- Vertex AI discontinuation (same IDs): Google's Vertex Imagen 4 model page lists a June 30, 2026 discontinuation date for imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 — roughly six weeks earlier than the Gemini API date.
- Platform split to track: Treat the Gemini API deprecations table as authoritative for ai.google.dev workloads. For Vertex or enterprise Agent Platform deployments, verify against Google's Cloud model pages and release notes — do not assume the August 17 date applies because a third-party tracker said so, or vice versa.
- Imagen 3 already retired: imagen-3.0-generate-002 shut down November 10, 2025 on the Gemini API, with imagen-4.0-generate-001 as the listed replacement at the time.
- Broader deprecation page: The same Gemini API page tracks 50+ model endpoints with shutdown dates through May 2028 — including Gemini 2.0 Flash (June 1, 2026), a late-2025 cluster of 2.5 preview models, Veo video endpoints (June 30, 2026), and embedding migrations — but Imagen 4 is the only family whose GA shutdown lands today.
Why it matters
Imagen 4 reached GA on the Gemini API barely 14 months before today's cutoff — an unusually short runway for a flagship image model. Google is consolidating image generation into the Gemini 3.x native image stack rather than maintaining a separate Imagen endpoint family on the developer API.
For teams still calling Imagen 4 through the Gemini API, requests to those model IDs should fail or return deprecation errors after shutdown. Workloads on Vertex that missed the June 30 discontinuation window may already be broken — or running on borrowed time if Google extended enforcement unevenly across surfaces.
The migration is also a code change, not a model-string swap. Firebase's Imagen-to-Gemini migration guide documents different SDK configuration (ImagenGenerationConfig → GenerationConfig with optional ImageConfig), different API entry points, and feature gaps — Gemini Image models always watermark with SynthID, return PNG only, and do not support multi-image batch generation the way Imagen did.
Unknown
- Which date wins on Vertex today? Vertex model pages still show June 30, 2026; Firebase's cross-platform migration doc states Imagen models "will shut down as early as August 17, 2026" across the Gemini Developer API and Agent Platform Gemini API (formerly Vertex AI). Google has not published a single reconciled timeline. If your pipeline runs on Vertex, check the Vertex model page and Cloud release notes — not third-party trackers alone.
- Ultra → which Gemini Image model? The Gemini API deprecations table lists gemini-3.1-flash-image for all three Imagen 4 GA IDs. Firebase's migration table maps imagen-4.0-ultra-generate-001 specifically to gemini-3-pro-image with different thinking-level settings for standard and fast variants. Confirm the mapping for your SDK and surface before migrating production quality tiers.
- Exact shutdown moment: Google lists August 17 as the shutdown date but does not specify UTC cutoff time on the deprecations page. Plan for full-day unavailability and test against the replacement model in staging beforehand.
Our take
Google's dual-surface documentation is the story here as much as the model retirement itself. A developer reading a third-party lifecycle aggregator sees June 30; a developer on ai.google.dev sees August 17 — and both can be "correct" depending on which Google page they trust. Until Google reconciles Vertex and Gemini API calendars in one authoritative place, the safe move is to verify against the page that matches your actual API endpoint, migrate early, and treat "Nano Banana" as a new integration — not a rename.
