Google has expanded its artificial intelligence family with the release of 3 new Gemini models. This update introduces Gemini 3.6 Flash, Gemini 3.5 Flash Lite, and a specialized security tool called Gemini 3.5 Flash Cyber. While 2 of the models are already available for public use, the cybersecurity version remains restricted to government agencies and trusted partners.
According to details on the Google developer portal, the main update is Gemini 3.6 Flash. This version acts as the primary workhorse for coding and complex data analysis. Tests run by the Artificial Analysis Index show that the model cuts output token usage by 17% compared to the older 3.5 version. In certain programming benchmarks, that reduction reaches up to 65%. Google priced the model at $1.50 per 1,000,000 input tokens and $7.50 per 1,000,000 output tokens to help developers save money on complex automated tasks.
The drop in token usage does not mean a dip in quality. Far from it. The model actually outperformed its predecessor in key benchmark tests. In the machine learning evaluation MLE Bench, its accuracy score jumped from 49.7% to 63.9%. Computer use test results in OSWorld Verified also grew from 78.4% to 83%. This means the model requires fewer steps and fewer tool calls to finish multi step workflows.
For tasks requiring immediate responses, Google released Gemini 3.5 Flash Lite. The company claims it is the fastest model in the current generation, pushing out 350 tokens per second. It is built for high volume workloads like document parsing and real world search queries. The price is set at $0.30 per 1,000,000 input tokens and $2.50 per 1,000,000 output tokens. That makes it incredibly cheap. In coding tests like Terminal Bench 2.1, its score jumped from 31% to 54% when compared to previous generations.
The final release is Gemini 3.5 Flash Cyber, which operates inside the CodeMender security platform. This model is designed specifically to scan and repair software vulnerabilities before attackers can find them. Because this technology has dual use potential, Google is limiting access to a private pilot program. Looking ahead, the company is already testing Gemini 3.5 Pro and has started training its next generation model, Gemini 4.
