Gemini 3.8 Flash: Faster AI for Coding & Cyber

Google has introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, expanding its Gemini model family with a stronger focus on AI agents, software engineering, advanced reasoning, and cybersecurity.

Announced on September 2, 2026, Gemini 3.8 arrives only three weeks after Gemini 3.7 Flash and represents Google’s third Flash release in six weeks. Google describes Gemini 3.8 Flash as its most intelligent workhorse model yet, designed to deliver stronger reasoning and coding capabilities while retaining the speed and relatively low cost associated with the Flash family.

The release is particularly significant because Google is introducing two closely related models for different environments: Gemini 3.8 Flash for general agentic and enterprise workloads, and Gemini 3.8 Flash Cyber for specialized defensive cybersecurity operations.

What Is Gemini 3.8 Flash?

Gemini 3.8 Flash is Google’s latest general-purpose Flash model, optimized for software development, autonomous AI agents, multi-step reasoning, and complex knowledge work.

Rather than simply generating a response from a single reasoning pass, the model is designed to perform more iterative work when a task requires it. Google says Gemini 3.8 Flash can take additional reasoning steps and repeatedly use tools to improve its results on difficult problems.

Google DeepMind’s model documentation also highlights configurable effort levels, allowing developers to balance quality, latency, and cost depending on the application.

This makes the model particularly relevant for AI systems that need to complete longer workflows rather than simply answer individual prompts.

For example, an enterprise AI agent may need to analyze information, call external tools, validate results, modify code, check the output, and then continue working. These longer workflows are becoming increasingly important as organizations move from traditional chatbots toward agentic AI systems capable of taking multiple coordinated actions.

Gemini 3.8 Flash Targets Long-Horizon Coding

Software engineering is one of the major areas Google has focused on with Gemini 3.8 Flash.

On the DeepSWE v1.1 long-horizon software engineering benchmark, Google reports that Gemini 3.8 Flash outperformed most larger frontier models when autonomously working through complex engineering problems from beginning to end.

The distinction between short coding tasks and long-horizon engineering matters. Generating a function or explaining an error is relatively contained. Real software engineering often involves navigating repositories, understanding dependencies, changing several files, running tools, investigating failures, and refining the implementation.

Models that can reliably maintain context and reasoning across these workflows could become increasingly useful inside AI coding assistants and autonomous development agents.

Google has demonstrated Gemini 3.8 Flash building interactive software experiences as well, including a playable DOS-style Google Maps application and an interactive 3D hardware visualization experience.

Stronger Reasoning for Enterprise AI Agents

Gemini 3.8 Flash is not limited to programming.

Google is positioning the model for specialized professional tasks that require structured analysis and multi-step reasoning.

The company reports improvements over Gemini 3.7 Flash across finance and legal agent evaluations. Gemini 3.8 Flash also achieved 54.9% on HLE-Verified, a benchmark covering difficult questions across STEM, humanities, and professional disciplines.

For businesses, this points toward a broader shift in enterprise AI.

Organizations are increasingly looking beyond assistants that summarize documents or draft text. The next stage involves agents that can reason about information, interact with enterprise systems, execute multi-stage workflows, and return validated outcomes.

Gemini 3.8 Flash appears designed with that type of AI architecture in mind.

What Is Gemini 3.8 Flash Cyber?

Alongside the general model, Google introduced Gemini 3.8 Flash Cyber, a cybersecurity-focused version designed primarily for defensive use.

Its core capabilities include autonomous vulnerability discovery and automated software patching.

On Google’s internal vulnerability evaluation covering complex codebases across 20 programming languages, Gemini 3.8 Flash Cyber achieved a success rate above 70%. Google also reported strong performance on CyberGym, an industry benchmark for autonomous vulnerability discovery.

The model also performed competitively on CWE-Bench, which evaluates whether AI systems can generate effective patches for software vulnerabilities. Gemini 3.8 Flash Cyber reached a 47.2% pass@1 score, close to a leading frontier model at 47.8%, while Google says it operates at substantially lower cost.

From Finding Vulnerabilities to Fixing Them

One important part of Google’s approach is its emphasis on defensive remediation.

Finding a vulnerability is useful, but organizations still need to understand, verify, fix, test, and deploy the correction.

Google’s new Fairwind Program combines Gemini 3.8 Flash Cyber with its CodeMender technology to help trusted organizations automatically identify and remediate vulnerabilities at agentic scale.

Rather than limiting AI to vulnerability detection, the system can generate and validate potential code fixes within secure environments. Google says the initiative is initially aimed at government organizations, critical infrastructure operators, software maintainers, and other trusted partners.

Google has already been expanding this approach inside Chrome. Its security teams use multi-agent workflows in which AI agents propose fixes, critic agents evaluate them, and additional agents help create tests before human review.

Real-World Cybersecurity Results

Google has also shared practical examples of Gemini 3.8 Flash Cyber being used in security environments.

According to Google, the Chrome Security team found that the model generated 2.6 times more correct vulnerability patches than the best larger commercial models evaluated by the team.

Google also reports that Wiz observed between 7.5% and 9.7% higher recall on an internal penetration-testing benchmark while operating at between 2.3 and 5.2 times lower cost than other leading frontier models.

In another test, Google’s Cloud Vulnerability Research team used Gemini 3.8 Flash Cyber to identify a critical foundational vulnerability in less than two hours, a type of discovery Google says could traditionally require months of research.

These results suggest AI’s cybersecurity role is moving beyond simply helping analysts interpret alerts. Increasingly capable models may participate directly in vulnerability discovery, remediation, testing, and security engineering.

How Much Does Gemini 3.8 Flash Cost?

Google launched Gemini 3.8 Flash at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, matching Gemini 3.7 Flash’s introductory pricing.

That introductory rate is scheduled to remain available until December 31, 2026. Google states that beginning January 1, 2027, pricing will increase to $1.50 per million input tokens and $7.50 per million output tokens.

The model’s adjustable effort levels are also important here. Applications that do not require maximum reasoning depth can use lower effort settings to reduce token usage.

Where Is Gemini 3.8 Flash Available?

Gemini 3.8 Flash is being distributed across Google’s AI ecosystem.

Developers can access it through the Gemini API and Google AI Studio, while enterprises can use it through Gemini Enterprise. Google AI Pro and Ultra subscribers can also access the model across the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.

Gemini 3.8 Flash Cyber has a more restricted release because of its advanced cybersecurity capabilities. Access is being provided to trusted defenders through Google’s Fairwind Program.

Why Gemini 3.8 Flash Matters

Gemini 3.8 Flash reflects a wider change happening across enterprise AI.

The focus is increasingly shifting from asking an AI model a question to giving an AI system a goal and allowing it to reason, use tools, evaluate results, and continue working toward an outcome.

For developers, that could mean more capable coding agents. For enterprises, it could mean AI-driven workflows that operate across applications and business data. For security teams, Gemini 3.8 Flash Cyber demonstrates how specialized AI agents could help find and remediate vulnerabilities considerably faster.

Organizations exploring these capabilities will still need strong governance, secure integrations, human oversight, and well-designed workflows around the underlying models. Teams such as Codimite, which works across AI development, agentic automation, enterprise AI integration, and AI-augmented software engineering, are operating in this broader space as businesses evaluate how models such as Gemini can be applied safely to real operational workflows.

Gemini 3.8 Flash therefore represents more than another incremental model update. It is another indication that the competition in AI is moving toward long-running agents, stronger reasoning, specialized models, and AI systems capable of completing meaningful work across increasingly complex environments.