Gemini 3.1 Pro: The Deprecation Nobody Announced Loudly Gemini 3.1 Pro on Vertex AI shipped on February 19, 2026, in public preview. If y...
Gemini 3.1 Pro: The Deprecation Nobody Announced Loudly
Gemini 3.1 Pro on Vertex AI shipped on February 19, 2026, in public preview. If you have any Vertex AI integrations pointing to gemini-3-pro-preview, check them now. That model was discontinued on March 26. Google doesn’t always surface these deprecations loudly, and a missed swap will break things quietly.
The replacement, gemini-3.1-pro-preview, handles the same workloads but with better reasoning, a 1M token context window, and a new MEDIUM thinking level parameter that sits between fast and deep. That middle setting is genuinely useful. Before this, you picked between quick-and-shallow or slow-and-expensive, with nothing in between.
There’s also a gemini-3.1-pro-preview-customtools variant released on February 23, optimized for bash-heavy and custom tool workflows. Worth noting: Google explicitly warns it may degrade quality for non-agentic use cases. It’s not a drop-in for general use. I appreciate that they said it outright instead of burying it.
Firestore is Finally Worth Re-evaluating
The bigger shift, at least for anyone doing serious data work on Google Cloud, is what happened to Firestore. The Enterprise edition now supports pipeline operations, and this is the change Firestore developers have been hacking around for years. You can now chain query stages server-side: unnest arrays, aggregate, filter on aggregated results. Previously you’d maintain separate metadata collections just to count tag occurrences or build frequency maps because Firestore couldn’t do any of that inline. Pipeline mode changes the calculus on what you even need a separate analytics layer for.
The migration path from Standard to Enterprise isn’t painless though. It’s a manual export, import, and full index and security rule recreation. No in-place upgrade path yet, which is a real cost on active projects. Also: no emulator support in Pipeline mode, no realtime listeners, and vector search performance in preview still lags the Standard edition. I wouldn’t push this to production before Next.
BigQuery and Apigee: Quieter but Useful
The BigQuery Studio Gemini assistant also picked up real capabilities in early 2026, not just surface UX polish. The Conversational Analytics API now lets you query BigQuery data through natural language without writing SQL. The assistant can schedule jobs, explain failed query root causes, and help with resource discovery. AI functions for text and image processing in BigQuery hit GA as well. Whether your team will actually let analysts skip SQL entirely is a different conversation, but the tooling is there.
Apigee got a few things worth tracking too. The AI-powered spec boost add-on, currently in public preview, automatically enriches API specs so LLM-based agents can consume them more reliably. Kubernetes Secrets support in Apigee hybrid also landed, which matters if you’re running GitOps pipelines in compliance-heavy environments.
Google Cloud Next 2026 is April 22-24 in Las Vegas. The expected focus areas are agentic AI in production and AI Hypercomputer infrastructure. Nothing confirmed on specific launches yet, but if Firestore pipelines and Gemini 3.1 are what preview looks like, the GA announcements at Next could be worth paying attention to. We’ll see.
Need help navigating any of this on a live project? Talk to the ATXSoft team.
References
- What’s new with Google Cloud
- Gemini 3.1 Pro on Vertex AI – Google Cloud Docs
- BigQuery Studio Gemini Assistant – Google Cloud Blog
- Google Cloud Next 2026 – Official Event Page
