Google Ships Gemini 3.6 Flash and a Cybersecurity-Tuned Model as 3.5 Pro Languishes

Google today rolled out three new Gemini models, headlined by a faster, cheaper successor to the Gemini 3.5 Flash that debuted at I/O just months ago. The company also introduced a specialized cybersecurity model and a lightweight Flash Lite, but offered only a cryptic update on the flagship Gemini 3.5 Pro, which was supposed to launch in June.

Gemini 3.6 Flash replaces 3.5 Flash as the default mid-range model in the Gemini API and app. Google claims it’s “marginally more capable” across the board, with a notable leap in coding: the DeepSWE test score jumps to 49% from 3.5 Flash’s 37%. Computer-use capabilities, now standard in the API, improved to 83% on OSWorld from 78.4%. More importantly for developers watching their budgets, the model uses about 17% fewer tokens per task and costs $1.50 per million input tokens and $7.50 per million output tokens—down from $1.50 and $9 respectively on the 3.5 version.

On the efficiency front, Gemini 3.5 Flash Lite is Google’s fastest model to date, hitting 350 tokens per second. It’s positioned as a cost-effective backbone for scaling agentic systems, with pricing at $0.30/1M input and $2.50/1M output tokens. Benchmark performance trails frontier models from a year ago, but the speed-to-cost ratio makes it ideal for high-volume use cases like Google’s AI Overviews.

The most unusual addition is Gemini 3.5 Flash Cyber, Google’s first LLM tuned specifically for cybersecurity workflows. The company says it approaches the capabilities of much larger models like Claude Mythos in finding and fixing vulnerabilities, while retaining the efficiency of a Flash-class model. Acknowledging the dual-use risk, Google will release it only through a limited pilot in its CodeMender agent, available to trusted partners and governments.

Meanwhile, the fate of Gemini 3.5 Pro remains unclear. Google confirmed the model is still in testing with unnamed partners, with a release “as soon as it’s ready.” The company has also started pre-training for the next-generation Gemini 4, described as “more ambitious” than previous efforts. With the Pro model’s silence and the Flash line’s rapid iteration, Google appears to be prioritizing efficiency and specialization over raw scale, at least for now.