The new models focus on cost and efficiency, particularly the Gemini 3.6 Flash, which reduces token use by 17% and as much as 65% in some benchmarks. It is also much less pricey, coming in at $1.50 per million input tokens and $7.50 in output.
3.5 Flash-Lite is built for faster agents, improving on 3.0 Flash in some cases and ticking in at 350 output tokens per second. It is also competitively priced, at $0.30 per million input tokens and $2.50/M for output.
Finally, Google is launching their first cyber model. 3.5 Flash Cyber is made for finding security vulnerabilities. It can detect, validate and patch security issues «at scale,» Google says, and is offered at a much lower cost than «larger models.» It is only available to governments and «trusted partners» as a limited access pilot program, just like Anthropic’s Mythos and GPT Cyber.
Despite not launching a new top model, Google is reporting strong growth of it’s Gemini offerings, outgrowing every other business segment with 82%, and they say it is used by 90% of the Fortune 100 companies.
Read more: Google’s introduction, DeepMind on Cyber, Sundar Pichai’s X post. Writeups on 9to5Google and CNBC.