Google Announces Gemini 3.8 Flash and Defense-Specialized Cyber Model

Google stated that 3.8 Flash shows significant improvements over 3.7 Flash in software engineering, agent tasks, and multi-step inference. The company positions this as its "best inference and coding model to date."

As the reason for the performance increase, Google cited a design choice that allows the model to perform more computation. For complex tasks, the model employs a mechanism that executes additional inference steps and iteratively calls tools. Consequently, token consumption may increase at higher effort levels.

Pricing will remain at the same level as 3.7 Flash: $0.75 per 1 million input tokens and $3.75 per 1 million output tokens. This pricing will apply until 2026-12-31, after which it will transition to standard pricing.

Meanwhile, Gemini 3.8 Flash Cyber will not be available for general release. Google stated that it will be provided only to trusted defenders through the newly established "Fairwind Program." This limited program provides phased access to entities such as government agencies and critical infrastructure operators.

3.8 Flash Cyber is provided with relaxed cybersecurity safeguards. In contrast, 3.8 Flash includes built-in countermeasures against misuse in the areas of CBRN and cyberattacks.