Google’s latest Gemini models target complex coding, autonomous agents, advanced reasoning and cybersecurity while maintaining Flash-series speed and pricing.
Google has introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, two new artificial intelligence models designed to improve complex reasoning, software engineering, autonomous agents and cybersecurity.
Announced Sept. 2, the models build on Gemini 3.7 Flash, with Google describing Gemini 3.8 Flash as its most intelligent Flash model yet. The company said the new model delivers significant gains in software engineering, agentic tasks and multi-step reasoning while retaining the speed and cost advantages of the Flash series.
Gemini 3.8 Flash is available at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, matching the introductory pricing of Gemini 3.7 Flash. Google said the promotional pricing will remain in effect through Dec. 31, 2026, before increasing to $1.50 per million input tokens and $7.50 per million output tokens from Jan. 1, 2027.
A major focus of the model is long-horizon software engineering. Google said Gemini 3.8 Flash can autonomously tackle complex engineering problems from start to finish and outperformed most larger frontier models on the DeepSWE v1.1 benchmark.
The model also posted a 54.9% score on HLE-Verified, while Google reported strong results on finance and legal agent benchmarks. The company attributed the gains partly to the model’s ability to perform additional reasoning steps and use tools iteratively on complex tasks.
Developers can adjust the model’s effort levels to control token use, cost and performance. Google will also continue supporting Gemini 3.7 Flash for workloads where computing efficiency is the priority.
The second model, Gemini 3.8 Flash Cyber, is focused specifically on cybersecurity. Google said it delivers frontier-level performance in autonomous vulnerability discovery and automated patching and is available to trusted defenders through the company’s new Fairwind Program.
On an internal benchmark covering vulnerabilities across codebases written in 20 programming languages, Google said the model achieved a success rate above 70%. On the external CWE-Bench patching benchmark, it recorded a 47.2% pass@1 score, compared with 47.8% for a leading frontier model, while costing significantly less, according to Google.
Google is already using Gemini 3.8 Flash Cyber internally. The company said its Chrome Security team produced 2.6 times more correct vulnerability patches with the model than with the best commercial models tested. Google’s Cloud Vulnerability Research team also used it to identify a critical foundational vulnerability in less than two hours, compared with research that can normally take months.
Gemini 3.8 Flash includes safeguards against misuse involving CBRN and cyber offense, while the Cyber version uses more permissive cybersecurity mitigations for trusted defenders. Google said both models also show improved resistance to prompt-injection attacks.
About The Author
