Google announced Gemini 3.8 Flash and 3.8 Flash Cyber for agentic workflows and cybersecurity. Company posts and employee statements describe the models as delivering next-generation intelligence in those areas. Availability covers Gemini API, Google AI Studio, Android Studio, Antigravity, and Gemini Enterprise users. Internal teams noted gains over prior Flash versions for coding tasks. The official blog post presents the pair as updated offerings in the Gemini line.
Google broke the law, but breaking up Google was not necessarily the best answer for advertisers. A federal judge ruled last year that Google illegally monopolized key parts of the online advertising market. The Justice Department wanted Google to sell its AdX ad exchange as part of the remedy. This week, Judge Leonie Brinkema rejected that request, and I believe keeping Google intact is good for the advertising industry.
Advertisers want a fairer marketplace where Google does not use its control of different parts of the ad-tech business to favor itself. The decision to keep Google intact while imposing rules designed to make the market more competitive avoids the disruption that could come with breaking up Google’s ad-tech business.
This case was not primarily about whether advertisers were paying too much for ads. The court found that Google had illegally monopolized the publisher ad server and ad exchange markets. Google’s position across different parts of the ad-tech business gave it the ability to favor its own products and make it harder for competitors to gain ground.
A breakup would have attacked that problem, but it also would have introduced plenty of uncertainty into a digital advertising market that thousands of advertisers and publishers rely on every day. The court is taking a different approach: keep Google’s ad-tech business together, but put restrictions on how Google can use its market power.
Advertisers should be watching for whether competing ad-tech companies get a fairer opportunity to compete, whether Google faces meaningful limits on favoring its own products, and whether advertisers ultimately gain more choice without adding unnecessary complexity to media buying.
Google avoiding a breakup does not mean the antitrust case accomplished nothing. The ruling gives the court an opportunity to address Google’s anticompetitive practices without taking apart an ad-tech business that advertisers and publishers rely on every day.
Now we need to see whether the remedies actually make it easier for other ad-tech companies to compete with Google. If they do, advertisers should have more choices, and Google will have more pressure to earn their business.
Google has announced two new variants of its Gemini 3.8 Flash model: a standard "workhorse" model optimized for agentic tasks, software development, and multi-step reasoning, and **Flash Cyber**, a specialized model for autonomous vulnerability detection and mitigation. Marking Google’s third Flash release in six weeks, the update delivers significant performance leaps over 3.7 Flash at highly competitive pricing, while actively shifting the advantage back to cybersecurity defenders.
1. Gemini 3.8 Flash (Standard)
*The customizable "workhorse" for complex, multi-step workflows.*
- **Core Capabilities:** Excels in software engineering, agentic tasks, and specialized knowledge domains (e.g., finance, law).
- **Specs:** 1M-token input window, 64K-token output limit. Fully multimodal (text, images, audio, video, PDF).
- **Customizable Effort:** Developers can adjust model "diligence" to balance quality, cost, and latency (lower token overhead for efficiency, or higher effort for complex reasoning).
- **Key Benchmarks:**
- **Agent Arena:** Ranked #14 (surpassing DeepSeek-V4-Pro; 3.7 Flash was #32).
- **Text Arena:** Debuted at #7 (ahead of Claude Opus 5).
- **Specialized:** Outperformed frontier models on Vals Finance Agent V2, Harvey’s Legal Agent Benchmark, and scored 54.9% on Humanity’s Last Exam (HLE)-Verified.
- **Demonstrated Use Cases:** Generating a 3D puzzle game in Antigravity, a functional DOS-style Google Maps, and real-time topographic map visualizers from USGS datasets.
2. Gemini 3.8 Flash Cyber
*Google’s most capable cybersecurity model, prioritizing defense over offense.*
- **Core Capabilities:** Autonomous vulnerability discovery, automated patching, and high prompt-injection robustness.
- **Key Benchmarks:** 86.2% on CyberGym, 47.2% on CWE-Bench (AI patching), and >70% success rate discovering vulnerabilities across 20 programming languages (internal).
- **Real-World Validation:**
- **Google Chrome:** Produced 2.6x more correct patches for Chrome vulnerabilities than larger commercial models; recently uncovered a subtle, 13-year-old bug in Chromium.
- **Wiz (recently acquired by Google for $32B):** Achieved 7.5%–9.7% higher recall of real-world vulnerabilities at 2.3x–5.2x lower cost than leading frontier models.
- **Google Cloud Vulnerability Research:** Discovered a critical foundational vulnerability in under 2 hours (a process that typically takes months).
- **Availability:** Currently rolling out to "trusted defenders" (government, critical infrastructure) via Google’s **Fairwind Program**. Features permissive cybersecurity mitigations and strict safeguards against offensive misuse (e.g., CBRN).
3. Pricing & Availability
- **Pricing:** $0.75 per million input tokens / $3.75 per million output tokens (matching introductory 3.7 Flash pricing).
- **Access:** Available now in Gemini Enterprise. Developers can test via Google AI Studio, Google Antigravity, Android Studio, or Stitch (for UI generation).
- **Legacy Support:** Gemini 3.7 Flash remains fully supported for "efficiency-first" workloads.
4. Strategic Context
- **The "Vulnerability Apocalypse":** As noted by Chrome Engineering Director Doug Turner, generative AI has caused a "hockey stick" spike in reported software vulnerabilities. Attackers only need to find one flaw; defenders must find them all.
- **Google’s Strategy:** By heavily investing in vulnerability *fixing* over exploitation, and deploying Flash Cyber internally to secure its own codebase, Google is equipping defenders with expert-level, cost-effective AI agents to counter both human and AI-driven threats.