Gemini 3.6 Flash launches with Flash Cyber pilot

Google has released Gemini 3.6 Flash, added a cheaper Gemini 3.5 Flash-Lite tier and opened a limited pilot for Flash Cyber, a security-tuned model. The company also said Gemini 4 is now in pre-training. For developers, the announcement is less about a new flagship model today than a lower-priced Gemini stack for production workloads.

Price is the clearest change. Google said Gemini 3.6 Flash uses 17 per cent fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index. It set pricing at $US1.50 (about $2.30) per 1 million input tokens and $US7.50 (about $11.40) per 1 million output tokens. Gemini 3.5 Flash-Lite is cheaper again at $US0.30 (about 46 cents) per 1 million input tokens, and Google said the lighter model can generate up to 350 tokens a second.

Google is putting Flash 3.6 in the heavier production slot. Flash-Lite sits beneath it for cheaper routing, summarisation and other high-volume inference jobs. The distinction is dry, but it is the kind of change that shows up in monthly bills for teams running support bots, coding helpers or document workflows at scale.

“Our newest Gemini models deliver the efficiency, latency, and reliability to build AI agents at scale,” Tulsee Doshi, Google’s senior director of product management, said.

Google’s developer documentation describes Gemini 3.6 Flash as the general-purpose production tier, with Flash-Lite positioned as the throughput option. For Australian software teams already testing Gemini inside internal tools, lower output costs can widen the set of jobs worth automating, especially where a bot generates long responses or runs the same task thousands of times a day.

Flash Cyber is narrower and harder to access. In a separate Google DeepMind post, the company said the model will enter a limited-access CodeMender pilot for governments and trusted partners, then expand over time. DeepMind is pitching it at defensive security workflows, not everyday developer use.

Most enterprise buyers still cannot test that model.

That limit is important because the cyber model is the most distinctive part of the release. For local teams, the deployable change this week is cheaper Flash pricing and a faster Lite tier, not a specialist model they can immediately plug into a security operations centre.

The release also moves Google’s Gemini story back to shipped software after recent attention on models still in testing. It is a deployment story first, with the cyber work parked behind a partner gate.

Google also said it had started its “most ambitious pre-training run yet” for Gemini 4. As Ars Technica noted, that leaves the next flagship as a teaser without a ship date, while the models landing now are the lower-cost Flash options built for high-volume AI agents.