What's this about?
Google released three new Gemini models at once. The headline model, 3.6 Flash, does higher-quality work than its predecessor while using fewer tokens and costing less ($7.50 instead of $9 per million output tokens). Flash-Lite is a cheaper model for repetitive bulk work like sorting tickets, and Flash Cyber is a small model specialized in finding and patching security vulnerabilities.
What exactly is a "Flash" model?
Model families come in tiers, like coffee sizes. The big flagship models are the smartest but slow and expensive; "Flash" is Google's mid-tier — fast and cheap enough to run millions of times a day, for tasks that don't need a genius. Most real production AI traffic (summarizing, extracting, routing) runs on this tier, which is why a price-and-efficiency upgrade here matters more to businesses than a new flagship.
What's the community saying?
- Skeptics say the benchmarks are unimpressive and Google only compared against its own previous models, not competitors.
- Several note GLM-5.2 (Chinese, open-weights) is both smarter and cheaper — the Flash tier is being squeezed by open models.
- Confusion over inconsistent benchmark numbers in the announcement (65% vs 49% on the same benchmark).
- Others see an encouraging faster release cadence from Google, and hope for a 3.5 Pro with its multimodal strengths.