Google ships Gemini 3.6 Flash, Flash-Lite, and a cyber model
Google's new Flash tier targets agentic workloads with fewer output tokens and lower prices, plus a cybersecurity model piloted with governments.
Google introduced three Gemini models on July 21 — Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber — a coordinated refresh of its mid-tier lineup aimed squarely at production agent workloads, while its frontier 3.5 Pro remains in testing.
Efficiency as the headline
Gemini 3.6 Flash is positioned as the new workhorse. Google says it delivers stronger coding and computer-use performance than 3.5 Flash — 49% versus 37% on DeepSWE and 83.0% versus 78.4% on OSWorld-Verified — while emitting about 17% fewer output tokens per task, per Artificial Analysis measurements. Since agentic systems pay for every intermediate token, that efficiency claim is as commercially significant as the raw scores. Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens.
Gemini 3.5 Flash-Lite targets high-volume, repetitive jobs such as ticket sorting and data extraction, streaming around 350 output tokens per second at $0.30/$2.50 per million tokens. Google cites a jump from 31% to 54% on Terminal-Bench 2.1 over the prior Lite generation.
A security specialist
The third release, Gemini 3.5 Flash Cyber, is a cybersecurity-tuned model paired with Google's CodeMender agent for finding and patching vulnerabilities. Google says it approaches frontier-model performance on the CyberGym benchmark at Flash-tier cost, and is initially limited to a pilot for governments and vetted partners — a distribution choice that mirrors the industry's growing caution around offensive-capable models.
Both general-purpose models are rolling out immediately in the Gemini API, AI Studio, and the Gemini app. Google confirmed Gemini 3.5 Pro is still being tested with partners and that Gemini 4 pre-training is underway.
Why it matters
The economics of AI agents are decided in the mid-tier, where volume lives. By competing on tokens-per-task rather than benchmark ceilings — and by shipping a gated security model the same week rivals face scrutiny over cyber-capable systems — Google is betting that efficiency and controlled distribution, not just raw capability, will define the next phase of the model race.