Google's DeepMind Releases Gemini AI Model with Enhanced Performance and Safety Features
Google's DeepMind has released two new variants of its Gemini AI model: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber.
Despite sharing the same foundational intelligence, these models differ in their safety envelopes and access controls.
Gemini 3.8 Flash is generally available for deployment through various APIs and platforms, including Google AI Studio and Android Studio, at a cost of $0.75 to $3.75 per million tokens until December 31, 2026.
The model has been fine-tuned to execute more reasoning steps and tool calls on complex tasks, but this comes at the cost of higher token consumption and reduced compute efficiency.
Google's own developer guide warns that users should consider staying with the previous version, Gemini 3.7 Flash, if compute efficiency is a major concern.
Gemini 3.8 Flash Cyber, on the other hand, is not openly deployable and requires access through the Fairwind Program for vetted defenders such as government authorities and software maintainers.
The model has shown significant gains in performance on various benchmarks, including DeepSWE v1.1, Vals Finance Agent V2, and Harvey's Legal Agent Benchmark, but its high cost and limited availability have raised concerns about accessibility.