Gemini Flash splits into three models for cheaper agent workloads

Google DeepMind is breaking the Gemini Flash line into more specialized production choices: Gemini 3.6 Flash for higher-quality agent work, Gemini 3.5 Flash-Lite for high-throughput tasks, and Gemini 3.5 Flash Cyber for security workflows inside CodeMender.

The tweet framed the release as “three new models” for agents that are “faster, smarter, and cheaper.”

The Google DeepMind account usually posts Gemini updates, research, robotics work, and responsible AI material. This July 21 post had roughly 1.69 million views when fetched. The linked Google blog adds the numbers that make the release material for builders: 3.6 Flash uses 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index, and Google says it observed up to a 65% reduction on some DeepSWE tasks.

Pricing is the other signal. Google lists 3.6 Flash at $1.50 per million input tokens and $7.50 per million output tokens. Flash-Lite is pitched for scale: 350 output tokens per second, $0.30 per million input tokens, and $2.50 per million output tokens. Google also claims clear gains over 3.1 Flash-Lite, including 54% versus 31% on Terminal-Bench 2.1 and 72.2% versus 60.1% on GDM-MRCR v2.

The cyber model is narrower by design. Gemini 3.5 Flash Cyber is tuned for finding and fixing vulnerabilities and will be paired with CodeMender in a limited-access pilot for governments and trusted partners. That restriction reflects the dual-use nature of automated vulnerability discovery. What to watch next is whether these models reduce full workflow cost, not just token price. Agent systems spend money through repeated calls, tool use, retries, and verbose outputs, so token efficiency and latency will matter as much as headline benchmark scores. The source tweet is here.