Google DeepMind releases Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber — still no 3.5 Pro

Per Google's DeepMind blog and TechCrunch, Gemini 3.6 Flash is the workhorse of the release, cutting token usage by up to 17% while improving coding and multimodal benchmarks; pricing was reported around $1.50 per million input / $7.50 per million output tokens, with 3.5 Flash-Lite far cheaper at roughly $0.30 / $2.50. The 3.5 Flash Cyber model is aimed at cybersecurity researchers and red teamers through a limited-access pilot tied to CodeMender.

Jeff Dean amplified the efficiency angle on X with a side-by-side token demo, and the release drew heavy Hacker News discussion. But the conspicuous omission — Gemini 3.5 Pro, whose GA has slipped repeatedly — became the story for skeptics who see a pattern of shipping Flash-tier refreshes while the flagship stays delayed.

Context: TechCrunch also reported Google is developing an internal efficiency-focused AI chip dubbed 'Frozen v2,' slated for 2028 and claimed to be 6-10x more power-efficient per token, underscoring Google's dual push on model capability and the compute/energy stack beneath it. Logan Kilpatrick separately teased that Google has begun its 'most ambitious pre-training run yet, for Gemini 4.'

The cyber variant lands in a loaded week: with frontier guardrails blocking defenders during the Hugging Face incident, a Google model explicitly built for security work — but gated behind a pilot — invites the obvious question of who gets access and under what refusal policy.