Google ships three cheaper Gemini models as flagship slips

TL;DR:

  • Google DeepMind released three lightweight Gemini models on Tuesday: 3.6 Flash, 3.5 Flash-Lite and a cybersecurity-tuned 3.5 Flash Cyber.
  • The workhorse 3.6 Flash cuts output token usage by around 17% and is priced below its predecessor at $1.50 per million input tokens and $7.50 per million output.
  • Absent again was the flagship Gemini 3.5 Pro, delayed since a planned June launch after reportedly missing internal coding targets.

The releases lean hard on cost and efficiency rather than raw capability — Google’s pitch to firms scaling AI agents into production, where price per task matters more than benchmark headlines. On Google’s figures, 3.6 Flash improves coding and knowledge-work performance while taking fewer reasoning steps, and 3.5 Flash-Lite runs at 350 output tokens per second for high-throughput jobs like document processing.

The cyber variant is the most restricted. Built on 3.5 Flash and paired with Google’s CodeMender agent to find and patch vulnerabilities, it will be available only to governments and trusted partners in a limited pilot — an echo of the dual-use caution now standard across the labs, and pointed given that frontier models are already demonstrating offensive capabilities in the wild.

The flagship is the story

What Google withheld matters more than what it shipped. Gemini 3.5 Pro is the model Wall Street is watching as a test of whether DeepMind can keep pace with Anthropic and OpenAI. Since the well-received Gemini 3 family last November — which prompted a “code red” inside OpenAI — rivals have pushed out GPT-5.6 and Anthropic’s Fable 5. Product lead Logan Kilpatrick said 3.5 Pro is testing with partners and should “land soon,” adding that the most ambitious pre-training run yet, for Gemini 4, has begun.

Looking forward

For UK buyers, cheaper Flash models are the practical near-term win: chief executive Sundar Pichai claims firms could save more than $1bn a year by shifting most workloads to Gemini. But a slipping flagship, blamed on coding shortfalls, is a reminder that the enterprise value increasingly sits in the top-tier reasoning models — and those remain the hardest to ship. Pichai faces questions on Alphabet’s earnings call this week.

Share this article