Did you know that we have LinkedIn, Instagram and X accounts
that you can follow?
In this issue:
🤿Deep Dive: Gemini 3.8 Flash Is built for cybersecurity
🤝Our Sponsor: A closer look at the FDE job market
⚡Quick Hits: The latest in AI, tech, and productivity
⚒Tool Snapshots: Tools for AI, no-code, and productivity
🤖AI Made This: Interesting and inspiring creations made with AI
🤿 DEEP DIVE
Gemini 3.8 Flash Works Harder Without Costing More
Google is moving quickly with its Flash models. Just 3 weeks after 3.7 Flash, it has introduced Gemini 3.8, its 3rd Flash release in 6 weeks.
The headline is straightforward - Google says 3.8 delivers better reasoning, coding and agentic performance while keeping the same introductory pricing as 3.7 Flash. But there’s another part to this release »> one version is specifically built for cybersecurity.
Image credit: Google (with edits)
A model designed to keep working
Gemini 3.8 comes in 2 variants. The general-purpose 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, while 3.8 Flash Cyber is aimed at trusted cybersecurity defenders.
For 3.8 Flash, much of the improvement is focused on longer, more complicated tasks. Google says it performs strongly on long-horizon software engineering, where a model has to autonomously work through complex engineering problems from beginning to end.
The same pattern appears outside coding. Google reports improvements over 3.7 Flash on benchmarks covering finance and legal work, while 3.8 Flash scored 54.9% on HLE-Verified, a benchmark involving multi-step reasoning across STEM, humanities and professional fields.
There’s an interesting detail behind those gains: 3.8 Flash works harder.
On complex problems, Google says the model performs additional reasoning steps and makes iterative tool calls rather than stopping earlier. That can improve performance, but it can also mean using more tokens, particularly at higher effort settings. Developers who care more about compute efficiency can lower the effort level or continue using 3.7 Flash.
Then there’s the cyber version
Gemini 3.8 Flash Cyber takes the same underlying intelligence in a more specialized direction.
On CyberGym, a benchmark for autonomous vulnerability discovery, Google says it reached frontier-level performance and beat both 3.5 Flash Cyber and significantly larger frontier models. On Google’s broader internal benchmark spanning 20 programming languages, its vulnerability-discovery success rate exceeded 70%.
Google has also emphasized fixing vulnerabilities rather than offensive exploitation. On the external CWE-Bench patching benchmark, 3.8 Flash Cyber recorded a 47.2% pass@1, compared with 47.8% for a leading frontier model.
And this isn’t limited to benchmarks. Google says Chrome’s security team saw 3.8 Flash Cyber produce 2.6x more correct vulnerability patches than the best commercial models it compared against. Google’s Cloud Vulnerability Research team also used it to find a critical foundational vulnerability in under two hours, a discovery the company says would usually take months of research.
There is an important restriction, though. Because the Cyber model uses more permissive cybersecurity safeguards, it isn’t being released broadly. Access is limited to trusted defenders through Google’s new Fairwind Program.
The quick take
Gemini 3.8 arrives just three weeks after 3.7 Flash, making it Google’s third Flash release in six weeks.
3.8 Flash keeps 3.7 Flash’s introductory pricing while improving coding, reasoning and agentic performance.
The model can spend more reasoning steps and tokens on difficult tasks to improve results.
3.8 Flash Cyber focuses heavily on vulnerability discovery and automated patching.
The Cyber version is restricted to trusted defenders through the Fairwind Program.
🤝 OUR SPONSOR
AI’s Next Bottleneck Is Deployment.
New models are easy to demo. Making them work inside real customer operations is not.
That is where forward deployed engineers come in: bridging code, customer context, and production outcomes.
The free State of FDE Jobs 2026 report maps the emerging labor market around this work.
⚡ QUICK HITS
The latest in AI, tech, and productivity worth knowing
OpenAI’s new reasoning trick has AI safety researchers worried - OpenAI’s Astra model will reportedly use “opaque recurrence,” which processes queries in loops and can leave fewer readable traces than conventional chain-of-thought reasoning. While Astra’s use is expected to be limited, researchers worry heavier use could make AI reasoning much harder to monitor.
⚒ TOOL SNAPSHOTS
Futuristic tools within AI, no-code, and productivity
🔌 Monid
Give your AI agent one key for thousands of APIs.
Why it’s useful: Handy for agents that need to pull data or perform tasks across areas like SEO, social media, markets, media generation, and on-chain data without separate subscriptions.
Turn almost any subject into a course you can interact with.
Why it’s useful: Useful for students who want structured lessons with visuals, quizzes, mindmaps, and an AI tutor, or want to generate a course around what they’re learning.
Change how websites work without building an extension yourself.
Why it’s useful: Lets you customize sites you regularly use by describing the change you want, then save and control that feature from the Sider Chrome extension.
🤖 AI MADE THIS
Interesting and inspiring creations made with AI
ℹ️ ABOUT US
The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.
We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.
Our mission is to empower individuals, boost their productivity, and future-proof their careers.
We read all your comments - please provide your feedback!
Did you like today's email?
What more do you want to see in this newsletter?
