Firebase AI Logic now supports Gemini 3.6 Flash and 3.5 Flash-Lite, cheaper than 3.5 Flash, but Google shipped no flagship model alongside them.
Firebase AI Logic apps can now call Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, the Flash-tier models Google shipped on July 21, 2026 alongside a security-focused variant, Gemini 3.5 Flash Cyber (Google, July 2026). Both new models are listed as supported in Firebase’s AI Logic docs, reachable through the same Gemini Developer API and Vertex AI Gemini API paths mobile and web apps already use (Firebase AI Logic docs).
What actually changed #
Gemini 3.6 Flash cuts output-token pricing to $7.50 per million, down from $9 for 3.5 Flash, while input stays at $1.50 per million. Google says it needs about 17% fewer output tokens than 3.5 Flash to finish the same tasks, and the knowledge cutoff moves forward to March 2026. On Google’s own published benchmarks, 3.6 Flash beats its predecessor across the board: 49% versus 37% on the DeepSWE coding test, 83.0% versus 78.4% on the OSWorld-Verified computer-use benchmark, and 1,421 versus 1,349 on GDPval-AA v2 for knowledge work (Google, July 2026).
Gemini 3.5 Flash-Lite is the cheaper, faster sibling: $0.30 per million input tokens, $2.50 per million output tokens, and roughly 350 output tokens per second, aimed at high-volume, latency-sensitive work like agentic search or document processing rather than raw capability. Both models keep the 1-million-token context window Gemini 3.x apps already expect.
Why it matters for Firebase apps #
AI Logic exists to let mobile and web clients call Gemini directly through a Firebase-hosted proxy, without an app carrying its own API key. That means every price cut and benchmark gain on the model side lands in Firebase apps the moment the docs list the model as supported, with no SDK changes required beyond pointing a request at the new model ID. For apps already built on 3.5 Flash, swapping in 3.6 Flash is a plausible free upgrade: same context window, lower output cost, higher scores on coding and tool-use benchmarks that matter for the agentic features AI Logic is increasingly built around.
What the flagship gap means #
Google shipped two Flash-tier models and a security variant, but no accompanying Gemini 3.5 Pro. That absence was the loudest thread in the Hacker News discussion of the release: commenters read it as a sign Google’s compute is stretched thin, pointing to sluggish, inconsistent coding sessions elsewhere in Google’s stack as circumstantial evidence, with some speculating a delayed Pro model is either too expensive to serve at scale or still working through alignment review (Hacker News, Hacker News). Other commenters pushed back, reporting that 3.6 Flash dispatches parallel tool calls more reliably than 3.5 Flash did, which matters more than raw capability for agent workloads that fire off several function calls at once. Firerun’s take: for Firebase developers, the practical read is narrower than the capacity debate. The Flash-tier upgrade costs nothing to adopt, but anyone waiting on a Pro-tier bump for AI Logic should not expect it on this release’s timeline.
Where this fits #
This follows Firebase’s pattern of adopting new Gemini releases into AI Logic within days rather than waiting for a dedicated Firebase announcement, the same fast-follow behavior seen when AI Logic picked up Gemini 3.x support at Google I/O 2026 (Firebase, May 2026) and again when it added Apple Foundation Models routing (Firerun, June 2026). Combined with App Check now being enforced by default on new AI Logic projects (Firerun, July 2026), Firebase’s AI Logic surface keeps getting cheaper and better-guarded in the same month, even without a flagship model to headline it.