Google said on September 2 it introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, expanding its AI models for coding, reasoning, and cybersecurity.

The company said Gemini 3.8 Flash matches the speed and low cost of Gemini 3.7 Flash while improving benchmark performance, including on DeepSWE, which measures long-horizon software engineering tasks, and HLE-Verified, which tests multi-step reasoning.

Google said Gemini 3.8 Flash Cyber achieved a 47.2% pass@1 score on CWE-Bench, an external benchmark for patching software vulnerabilities.

The model is available only to trusted defenders through the Fairwind Program.

Google described Fairwind as a proactive cyber-defense framework for governments and enterprises, while a DeepMind page said it provides trusted access for cyber defense.

The limited rollout mirrors a broader industry pattern. OpenAI and Anthropic have also limited access to cyber-capable models.

🔗 Source: Google

Recent Google developments