Google Cloud warns: Gemini 3.5 Flash leaving 'global' on Aug. 4

Google Cloud published release notes on July 27, 2026 that include an operational announcement affecting Gemini Flash models in the Gemini Enterprise app. The notes state that Gemini 3.5 Flash will be removed from the "global" region on August 4, 2026.

The line in the Gemini Enterprise release notes is explicit: the model will be removed "as a model from the `global` region in the Gemini Enterprise app." That phrasing indicates this change is about availability in the in‑app model selector rather than a blanket removal of the model everywhere.

The July 27 entry sits among a set of rapid updates this summer — Google made Gemini 3.6 Flash available in the global and US multi‑regions days earlier, while other administrative changes to feature controls appeared in June. The timing makes August 4 a tight migration window for shops that have standardized on the global endpoint.

Enterprises should treat the change as one that can affect region planning and latency guarantees. Agent Platform notes and model‑serving guidance show tuned and served models are sometimes restricted to specific regions or multi‑region endpoints, and moving endpoints can change round‑trip times and the surface area of any SLOs you publish.

Google’s Agent Platform release notes also reference the Gemini Online Inference API SLA and recent administrative corrections to SLA language, underscoring that endpoint location ties into formal service commitments. Teams that assign latency SLAs to products running on Gemini models will need to reassess those commitments when endpoints change.

Cost is another operational variable. The Agent Platform documentation calls out Provisioned Throughput for fine‑tuned Gemini models and warns tuned inference can incur higher quota burndown rates, which translate into higher bill lines or different provisioning needs. Moving from a global endpoint to region‑specific endpoints can change usage patterns and provisioned throughput demands.

The technical release notes contain concrete constraints that matter to migration plans: supervised and reinforcement fine‑tuning for Gemini 3.5 Flash has been restricted to us‑central1 and europe‑west4, and tuned model serving is limited to the US and EU multi‑region endpoints. Those restrictions shape where trained or tuned endpoints can run.

Google’s model lifecycle page gives related context: gemini‑3.5‑flash was released May 19, 2026 and has a retirement horizon listed as at least 12 months from release, which means removal from the "global" region is distinct from full retirement and can be part of normal endpoint reshuffling. The model‑versions guidance also includes migration steps enterprises should follow.

Administrators should note earlier feature toggle changes: Google removed the Gemini 3.5 Flash feature management toggle in June, and the model was enabled by default in many multi‑regions. That history means some organizations may already have 3.5 visible in their apps even if they made no change, and they'll have limited control over turning the model off.

Practical steps for engineering and ops teams include: inventory which projects call the `global` endpoint, update endpoint configuration to region or multi‑region endpoints you control, add multi‑region fallbacks, and run latency and functional tests before August 4. Google’s migration guidance recommends updating application endpoints, testing mission‑critical features, and deploying updates in a controlled rollout.

Product and procurement teams should recast cost forecasts for any workloads that use Provisioned Throughput or tuned inference, and discuss the window with Cloud account teams if you need multi‑region access or quota adjustments. The release notes explicitly reference access requests for newer Flash models in specific multi‑regions, which is an available path if you need an immediate replacement.

In short, the August 4 removal from the Gemini Enterprise app's `global` region is an operationally meaningful event for customers who rely on specific Flash variants. It is not, according to the notes, an immediate retirement of the model everywhere, but it does create a short period in which engineers, product managers, and finance teams should act to avoid surprises. The release notes and model lifecycle pages are the authoritative reference for next steps.