Firebase AI Logic adds Gemini 3.1 model support

Firebase AI Logic now supports Gemini 3.1 models and sets clear retirement dates for older model families.

Firebase has updated its AI Logic model list with Gemini 3.1 options, including Pro, Flash, and Flash-Lite. The same page also warns developers that Gemini 2.0 Flash and Gemini 2.0 Flash-Lite will shut down on June 1, 2026, while all Imagen models end on June 24, 2026.

What Firebase is actually telling developers

The practical message is simple: if your app uses Firebase AI Logic, the supported model set has moved forward, and the old defaults are on a timer. Firebase now points developers toward Gemini 3.1 Pro for heavier reasoning, Gemini 3.5 Flash for fast general work, and Gemini 3.1 Flash-Lite for budget-sensitive traffic.

That matters because Firebase AI Logic is the path many mobile and web teams use to call Gemini models from client apps without building a separate server integration first. The docs also keep the model naming rules in plain sight: stable Gemini 2.5 names do not use a three-digit suffix, and they do not have an auto-updated alias.

• Gemini 3.1 Pro: advanced intelligence and complex problem-solving
• Gemini 3.5 Flash: fast performance with lower cost pressure
• Gemini 3.1 Flash-Lite: high-volume workhorse for tighter budgets
• Gemini 2.5 Pro, Flash, and Flash-Lite: still listed for current use

The naming rules matter more than they look

Model names are where teams often trip up during migrations. Firebase is explicit that stable Gemini 2.5 model names do not include a three-digit suffix, and they do not get an auto-updated alias. If you hard-code the wrong string, you can end up pinned to an outdated preview or a model that no longer matches your expectations.

That warning is especially relevant for teams that rely on remote config, feature flags, or client-side model selection. A small naming mistake can turn into a production issue when a model family changes behavior, pricing, or availability.

“The current generation of Gemini models is the most capable and useful family we’ve ever built.” — Demis Hassabis, Google DeepMind, Google I/O 2025 keynote

The quote matters because it matches the direction Firebase is taking here. Google is concentrating support around newer Gemini families, while older image and chat lines are being phased out on a fixed schedule. If you are shipping apps that depend on model stability, that means your migration work should happen now, not in the last quarter before shutdown.

How the new lineup compares

Firebase’s list makes the tradeoffs easy to read. The 3.1 models are aimed at the newest capabilities, while the 2.5 line still covers a wide set of text, reasoning, and multimodal tasks. The 3.5 Flash model is the interesting middle ground: it is described as frontier-class performance at lower cost, which gives teams a faster option without jumping straight to the most expensive tier.

For image generation, the direction is even clearer. Firebase now points developers to Gemini image models, including Gemini 3 Pro Image, Gemini 3.1 Flash Image, and Gemini 2.5 Flash Image. At the same time, all Imagen models are marked deprecated and scheduled for shutdown on June 24, 2026.

• Gemini 3 Pro Image: built for professional asset production
• Gemini 3.1 Flash Image: tuned for speed and high-volume use
• Gemini 2.5 Flash Image: optimized for efficient image generation
• Imagen: deprecated, with migration guidance already in the docs

That split tells you where Firebase wants developers to go next. If your app uses image generation, the migration path is no longer optional planning work; it is a calendar item. If your app uses text or multimodal chat, the bigger question is whether you need the reasoning depth of Pro, the balance of Flash, or the cost profile of Flash-Lite.

What teams should do before the deadlines hit

The smartest move is to audit every model string in your app, then map each one to a supported replacement. Start with any Gemini 2.0 usage, then check image generation flows that still depend on Imagen. After that, verify whether your app depends on preview names or aliases that could change under you.

If you want a practical order of operations, use this checklist:

• Inventory every Firebase AI Logic call in mobile, web, and backend code
• Replace Gemini 2.0 Flash and Flash-Lite before June 1, 2026
• Move Imagen workloads to Gemini image models before June 24, 2026
• Test latency, output quality, and billing impact after each swap

One more detail is worth noting: Firebase also calls out that Gemini 2.5 and later preview models released after June 2025 are only available through the Vertex AI Gemini API, with exceptions for Gemini Live API models. That is a sign that Google is tightening the distribution path for newer previews, so app teams should pay close attention to which API surface they are actually using.

For developers, the takeaway is straightforward. Firebase AI Logic is moving to newer Gemini families, older models have fixed end dates, and the safest codebase is the one that treats model names like versioned dependencies. The next step is simple: search your app for model strings this week, not after the first shutdown notice forces the issue.