Google Launches Gemini 3.6 Flash, 3.5 Flash-Lite & 3.5 Flash Cyber

Google Launches Gemini 3.6 Flash: Google just dropped three new Gemini models, and the headline isn't raw power — it's efficiency. Built for developers scaling AI agents, these releases quietly…

By Raunak Saha

July 21, 2026

2 min read

Comments

Google Launches Gemini 3.6 Flash: Google just dropped three new Gemini models, and the headline isn’t raw power — it’s efficiency. Built for developers scaling AI agents, these releases quietly signal where the entire AI race is heading next.

What Google Just Shipped

Announced on Google’s official Keyword blog, the trio builds directly on Gemini 3.5 Flash. Each model targets a different pain point developers face when running AI agents in production: cost, speed, and security.

Google Launches Gemini 3.6 Flash, 3.5 Flash-Lite & 3.5 Flash Cyber

Gemini 3.6 Flash is the star here. It cuts output token usage by 17% versus 3.5 Flash while improving coding and multimodal performance, according to the Artificial Analysis Index. Meanwhile, 3.5 Flash-Lite claims the crown as Google’s fastest 3.5-class model yet, and 3.5 Flash Cyber pairs with Google’s CodeMender agent to hunt down security vulnerabilities before attackers do.

Model	Best For	Key Stat
Gemini 3.6 Flash	Coding, knowledge work	17% fewer output tokens than 3.5 Flash
Gemini 3.5 Flash-Lite	High-throughput, low-latency tasks	350 output tokens/sec
Gemini 3.5 Flash Cyber	Cybersecurity vulnerability detection	Frontier-level CyberGym scores

Why This Matters Beyond the Benchmarks

Pricing tells its own story. 3.6 Flash comes in at $1.50 per million input tokens and $7.50 per million output tokens — cheaper per task despite the performance bump. 3.5 Flash-Lite goes even further at just $0.3/$2.5 per million tokens, aimed squarely at developers running massive agentic workloads.

Both models are live now via Google AI Studio and the Gemini API, with 3.6 Flash also rolling into Google Antigravity and Gemini Enterprise. 3.5 Flash Cyber, however, stays restricted — Google is limiting it to governments and trusted partners through CodeMender, a deliberate move given the dual-use risk of a model built to find software vulnerabilities.

This launch fits into a much bigger AI storyline our readers have been following. If you’re tracking how quickly foundation models are evolving, check out our recent coverage on AI in gaming and esports and our breakdown of tech industry announcements shaping 2026.

The Bigger Picture

Google isn’t stopping here — Gemini 3.5 Pro is already in partner testing, and the company confirmed it has begun pre-training Gemini 4. The Flash series may be the “budget” tier, but it’s clearly where the real agentic AI race is being won.