Google unveils Gemini 3.8 Flash and Gemini 3.8 Flash Cyber
Google on Wednesday introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, the latest additions to its Gemini AI model lineup, with the company positioning the release as a significant step forward in software engineering, reasoning, and cybersecurity.
The launch comes three weeks after Gemini 3.7 Flash, marking Google's third Flash release in six weeks, as reported by both CNBC TV18 and The Times of India. The new models arrive with the same introductory pricing as their predecessor: $0.75 per million input tokens and $3.75 per million output tokens, according to Google.
Performance claims and benchmarks
Google said its standard Gemini 3.8 Flash model is designed for software engineering, agentic tasks, and multi-step reasoning in specialised domains. On the DeepSWE v1.1 benchmark, which measures long-horizon software engineering capabilities, the company said the model outperforms most larger frontier models in autonomously solving complex engineering problems end to end.
Google CEO Sundar Pichai described the release in a post on X, writing, "Today we're introducing a new 3.8 Flash model, our 3rd Flash release in just 6 wks. It delivers significant leaps from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning. On DeepSWE v1.1, it outperforms most larger frontier models in autonomously solving complex engineering problems end to end, at a fraction of the cost."
The model also outperformed Gemini 3.7 Flash and other frontier models on the Vals Finance Agent V2 and Harvey's Legal Agent Benchmark, according to Google. It scored 54.9% on HLE-Verified, a test of multi-step reasoning across STEM, humanities, and professional fields, the company said.
Gemini 3.8 Flash Cyber: focused on security
The second variant, Gemini 3.8 Flash Cyber, is tailored for defensive digital security teams and focuses on vulnerability detection and automated patching. Google said the model produced 2.6 times more correct patches for vulnerabilities than larger models in evaluations involving real Chrome security bugs.
Pichai said the cybersecurity-focused model scored 86.2% on the CyberGym industry benchmark and 47.2% on CWE-Bench for patching. He also said the model recorded a success rate of more than 70% in discovering vulnerabilities across 20 programming languages on Google's internal benchmark.
According to Pichai's statements as reported by CNBC TV18, on the CyberGym benchmark, Anthropic's Opus 5 scored 83.8% and GPT-5.6 Sol scored 83.6%, indicating Gemini 3.8 Flash Cyber's score surpassed both. The Times of India also reported on the model's benchmark performance, with the company emphasising that the Cyber version is designed to help security teams autonomously identify and fix vulnerabilities.
Fairwind programme
Google launched the Fairwind programme on Wednesday as a limited-access initiative for governments and trusted partners, as confirmed by both CNBC TV18 and The Times of India. The programme will give defenders access to advanced Gemini models to autonomously find and fix vulnerabilities across critical infrastructure, public services, and national security systems.
Pichai said the Cyber model will be available to trusted defenders through the programme, with government agencies and cybersecurity partners also receiving access. According to CNBC TV18, the initiative is aimed at helping defenders use the models to secure critical systems, with the company describing it as a way to provide advanced AI capabilities to those responsible for protecting national security.
The Times of India reported that Google is distributing the Cyber model to verified defenders through its newly unveiled Fairwind Program, with the company emphasising the development of the model for defensive security teams.