Google says its new Gemini 3.8 Flash model ‘works harder’ but might cost more

Google has continued its rapid release of Gemini models, this time introducing the Flash 3.8 and Flash 3.8 Cyber on Wednesday. Despite only being launched three weeks ago, Flash 3.8 often surpasses older models like Gemini 3.7 Flash, particularly in agentic coding and computer use tasks. The new Flash model will be available at a lower introductory price of $0.75/$3.75 per million input/output tokens, which will increase to $1.50/$7.50 on December 31, 2026.

Flash 3.8 is being positioned as Google's "workhorse model," but Flash 3.8 Cyber, described as Google's "most capable cybersecurity model," is also noteworthy. This model is exclusively available through Google's Fairwind program to trusted partners such as Accenture, CrowdStrike, and Wiz. By launching this program, Google aims to help defenders respond to threats faster and more efficiently, shrinking the time between flaw detection and patching.

Google's Flash 3.8 model has demonstrated superior performance in long-horizon coding and agent tasks, matching or even surpassing models like GPT-5.6 Sol from OpenAI and Claude Sonnet 5 and Opus 5 from Anthropic. However, it does require more computational resources, which could increase token usage. To address this, developers can choose between different reasoning modes.

Google's Flash 3.8 Cyber model shows "frontier-level performance in autonomous vulnerability discovery" and has outperformed previous models in C and C++ code benchmarks. It has also produced more accurate patches than leading commercial models and achieved higher recall rates in penetration testing benchmarks while incurring lower costs. Additionally, the model includes enhanced safety measures to prevent misuse in sensitive domains.