Google has rolled out a fresh batch of artificial-intelligence models led by Gemini 3.6 Flash, promising better performance at a lower cost for the developers who build apps on top of it. The launch happened on 21 July 2026.
Alongside 3.6 Flash, the company introduced a lighter, faster model called Gemini 3.5 Flash-Lite and a security-focused version, Gemini 3.5 Flash Cyber, designed to hunt for software vulnerabilities.
The Flash line is what Google calls its workhorse, the everyday model meant for high-volume, routine tasks rather than the heaviest reasoning. The new 3.6 Flash uses 17 percent fewer output tokens than its predecessor, which effectively makes it cheaper and quicker to run at scale.
Google also cut the price. The cost of output dropped from $9.00 to $7.50 per million tokens, while input stayed at $1.50 per million. Its knowledge cutoff, the point up to which it has learned about the world, advanced from January 2025 to March 2026.
On performance, the model scored 49 percent on the DeepSWE coding benchmark, up from 37 percent, while its ability to operate a computer improved from 78.4 to 83.0 percent on the OSWorld-Verified test. Google says it needs fewer reasoning steps and tool calls to complete multi-step jobs.
The budget option, Flash-Lite, runs at about 350 output tokens per second and is priced far lower at $0.30 for input and $2.50 for output per million tokens. The cybersecurity model, meanwhile, will be offered only to governments and trusted partners through a limited-access pilot.
Google added that Gemini 3.5 Pro will roll out more widely soon, and confirmed it has already begun pre-training the next generation, Gemini 4, signalling that the rapid pace of releases shows no sign of slowing.
Source: Android Authority.
Comments 0