Firebase AI Logic Now Supports Gemini 3.6 Flash and 3.5 Flash Lite Models for Flutter Apps

Firebase AI Logic has added support for the latest Gemini stable models — gemini-3.6-flash and gemini-3.5-flash-lite — marking a significant upgrade for Flutter developers building AI-powered mobile and web applications. The new models bring enhanced performance, lower latency, and improved reasoning capabilities, making them immediately attractive for production Flutter apps that rely on generative AI features.

The announcement confirms that both models are accessible from mobile and web applications through either the Gemini Developer API or the Vertex AI Gemini API. For Flutter developers using the Gemini Developer API, accessing these new models does not require upgrading to the Blaze pricing plan, meaning hobby projects and early-stage applications can continue using the free tier.

What the New Models Bring

Gemini 3.6 Flash represents the latest advancement in Google's Flash model family, designed for speed without sacrificing quality. For Flutter applications, this translates to faster response times in chat interfaces and more fluid streaming of generated content. The model shows particular strength in structured data extraction, making it well-suited for Flutter apps that parse and display information from natural language inputs.

The gemini-3.5-flash-lite model offers a compelling option for Flutter applications where resource efficiency is a primary concern. Mobile devices running on older hardware or in emerging markets benefit from the lighter computational footprint. Despite its smaller size, it delivers capable performance for text summarization, classification, and simple conversational interactions.

There are important migration considerations. The temperature, topP, and topK sampling parameters are now deprecated in both new models. Developers who need deterministic responses should instead define system instructions with explicit rules for their specific use case. Additionally, prefilling model turns in requests is no longer supported — if the last non-empty turn is a model turn, the API returns a 400 error.

• Sampling parameters deprecated: temperature, topP, and topK are now ignored. Use system instructions for deterministic output
• No prefilled model turns: The last non-empty turn must be from the user. Use system instructions or user messages to provide context
• Vertex AI location: When using Vertex AI, set the location parameter to global

The deprecation of sampling parameters reflects a broader shift in Google's model design philosophy. Rather than relying on sampling adjustments to control output variety, newer models follow system instructions more precisely. For Flutter applications that previously depended on temperature adjustments, the recommended approach is to include variety instructions directly in the system prompt.

For Flutter developers using Firebase AI Logic's prompt template system, integrating these new models is straightforward. In the Firebase console, update the model field in the prompt template to either gemini-3.6-flash or gemini-3.5-flash-lite. The built-in template testing tools allow immediate validation before deploying to production. Since templates are stored server-side, no client-side Dart code changes are needed.

Flutter applications using custom integration code rather than Firebase AI Logic will need to update model names in their Dart source code. The Gemini API client libraries for Dart have been updated to support these model names, but developers should verify they are using the latest version of their chosen client package.

Practical Benefits for Flutter Applications

During internal testing, gemini-3.6-flash demonstrated approximately 30 percent lower latency compared to gemini-2.5-flash for typical mobile AI workloads. The improved reasoning capabilities are valuable for customer support chatbots, educational tutoring apps, and AI-assisted development tools that require coherent multi-turn conversations.

The gemini-3.5-flash-lite model fills an important niche for Flutter applications deployed on lower-end devices or in regions with limited bandwidth. Flutter apps targeting emerging markets benefit from having a capable but lightweight model option that delivers AI features without compromising user experience.

Both models are fully compatible with Firebase AI Logic's grounding capabilities, including Grounding with Google Maps. Flutter developers building location-aware AI applications can combine improved reasoning with real-time data from Google's database of places, creating apps that understand both language and physical context.

In summary, the addition of gemini-3.6-flash and gemini-3.5-flash-lite to Firebase AI Logic represents a meaningful upgrade for Flutter developers. The improved performance and lower latency directly benefit end-user experiences, while the migration path — particularly for teams using prompt templates — is practical and well-documented. Flutter teams can adopt these new models with confidence and deliver better AI experiences today.