Google Unleashes Third Gemini Flash Model, with Improved Performance and New Cybersecurity Features

Google has released its third Gemini Flash model in just six weeks, and it's already outperforming its predecessor in several key areas. The new model, Gemini 3.8 Flash, is available at a price of $0.75 to $3.75 per million input/output tokens, although this introductory rate will expire on December 31, 2026, and increase to $1.50 to $7.50 after that.

In a significant development, Google has also launched a new program called Fairwind, which offers access to the Gemini Flash Cyber model, its most capable cybersecurity model with frontier-level performance in vulnerability detection and automated patching. This model is only available to a select number of 'trusted defenders', including companies like Accenture and CrowdStrike.

According to Google's benchmarks, Gemini 3.8 Flash often outperforms other models on agentic coding tasks and computer use, although it still lags behind in some areas such as GDPVal, which tests models on knowledge work tasks. The company notes that its new model is powered by the same foundational intelligence as before, but further accelerated by long-running agentic loops designed to recursively evaluate and refine the underlying models.

Google's comparison with other models also highlights the rapid progress being made in AI research, particularly in China, where some models are offering better price/performance ratios. However, on cybersecurity benchmarks such as CyberGym, Gemini Flash 3.8 Cyber delivers frontier-level performance in autonomous vulnerability discovery and significantly outperforms larger frontier models.

The company also stresses that its new model is far more robust when it comes to prompt injections and has (almost) class-leading results in the Gray Swan IPI benchmark.