{"id":40561,"date":"2026-09-02T23:01:33","date_gmt":"2026-09-02T16:01:33","guid":{"rendered":"https:\/\/dps.media\/gemini-3-8-flash\/"},"modified":"2026-09-02T23:01:33","modified_gmt":"2026-09-02T16:01:33","slug":"gemini-3-8-flash-2","status":"publish","type":"post","link":"https:\/\/dps.media\/en\/gemini-3-8-flash-2\/","title":{"rendered":"Gemini 3.8 Flash: A Quantum Leap in Autonomous Coding and Cybersecurity Defense"},"content":{"rendered":"<?xml encoding=\"utf-8\" ?><p>September 02, 2026, Google DeepMind officially released <a href=\"https:\/\/dps.media\/en\/gemini-3-8-flash-2\/\">Gemini 3.8 Flash<\/a> along with the dedicated cybersecurity variant Gemini 3.8 Flash Cyber. This marks the third Flash-series launch in just six weeks, following the strong momentum of version 3.7. Not only maintaining lightning-fast response times and an extremely low price ($0.75 per 1 million input tokens and $3.75 per 1 million output tokens), version 3.8 also delivers a major leap in long-horizon coding capabilities and autonomous AI agent operations.<\/p><p>According to <a href=\"https:\/\/blog.google\/innovation-and-ai\/models-and-research\/gemini-models\/3-8-flash-and-3-8-flash-cyber\/\" rel=\"nofollow noopener\" target=\"_blank\">the official announcement from Google DeepMind<\/a> led by Tulsee Doshi (Product Director) and Raluca Ada Popa (Head of Gemini Security), both models share the same next-generation reasoning core, trained through recursive self-evaluation cycles to thoroughly solve real-world engineering problems.<\/p><p><img decoding=\"async\" src=\"https:\/\/dps.media\/wp-content\/uploads\/2026\/09\/blog_image_3-1.webp\" alt=\"Gemini 3.8 Flash\" style=\"display:block; margin:20px auto; max-width:100%; height:auto;\" title=\"\"><\/p><h2>Gemini 3.8 Flash \u2013 The hard-working mechanism for complex problems<\/h2><p>Instead of racing to increase the number of parameters and pushing operating costs through the roof, Google chose to optimize performance by making the model work harder. When handling coding requests or multi-step problems, Gemini 3.8 Flash does not immediately guess the answer. The model actively performs additional hidden reasoning steps while continuously calling and checking external tools to cross-verify results before reaching a final conclusion.<\/p><p>Real-world testing on international benchmark suites shows outstanding performance:<\/p><ul>\n<li><strong>Long-horizon software engineering (<code class=\"notranslate no-translate\" data-no-translation=\"\">DeepSWE v1.1<\/code>):<\/strong> Outperforming most expensive large Frontier models in autonomously detecting bugs, writing code, and fixing issues end-to-end at a fraction of the cost.<\/li>\n<li><strong>Specialized enterprise operations:<\/strong> Leading in financial tasks (<code class=\"notranslate no-translate\" data-no-translation=\"\">Vals Finance Agent V2<\/code>) and legal analysis (<code class=\"notranslate no-translate\" data-no-translation=\"\">Harvey's Legal Agent Benchmark<\/code>), demonstrating markedly greater stability compared with version 3.7.<\/li>\n<li><strong>Multidisciplinary reasoning:<\/strong> Scoring 54.9% on the benchmark scale <code class=\"notranslate no-translate\" data-no-translation=\"\">HLE-Verified<\/code>, demonstrating high accuracy across STEM and academic fields.<\/li>\n<\/ul><p><img decoding=\"async\" src=\"https:\/\/dps.media\/wp-content\/uploads\/2026\/09\/blog_image_4.webp\" alt=\"Gemini 3.8 Flash leads the DeepSWE v1.1 software engineering benchmark\" style=\"display:block; margin:20px auto; max-width:100%; height:auto;\" title=\"\"><\/p><p>Notably, developers can customize the level of effort (Effort Level). When running simple tasks, you can lower the reasoning level to reduce token consumption. But when encountering a difficult architectural bug, increasing the Effort Level allows the model to devote more resources to deep analysis and find the optimal fix.<\/p><h2>Real-world experiments: From 3D games to DOS maps<\/h2><p>The model's ability to turn ideas into software products is vividly demonstrated by Google through practical applications:<\/p><ul>\n<li><strong>Wizarding world 3D game:<\/strong> In the Google Antigravity environment, the model autonomously builds a complete 3D level using just a single iterative command. All puzzles, storyline, and graphical textures are smoothly created by AI in combination with Nano Banana.<\/li>\n<li><strong>DOS version of Google Maps:<\/strong> Fully recreating a classic DOS-style map through a single prompt, supporting location search, directions, and simulated Street View mode.<\/li>\n<li><strong>Real-time 3D topographic map:<\/strong> Directly integrating data from the United States Geological Survey (USGS) to build an interactive terrain model, supporting cross-sectional views and intuitive geological parameter visualization.<\/li>\n<li><strong>Hardware breakdown (<code class=\"notranslate no-translate\" data-no-translation=\"\">Hardware Anatomy<\/code>):<\/strong> In <a href=\"https:\/\/aistudio.google.com\/\" rel=\"nofollow noopener\" target=\"_blank\">Google AI Studio<\/a>, the model creates 3D tools using <code class=\"notranslate no-translate\" data-no-translation=\"\">Three.js<\/code>, automatically decomposing electronic device components according to precise physical proportions.<\/li>\n<\/ul><h2>Gemini 3.8 Flash Cyber \u2013 A breakthrough in automated cybersecurity defense<\/h2><p>The Gemini 3.8 Flash Cyber variant is designed as an intelligent shield specifically for cybersecurity defenders through the Fairwind Program. Google prioritizes vulnerability detection and automated patch generation rather than focusing on offensive weapons.<\/p><p>Experimental figures from leading security organizations demonstrate clear strength:<\/p><ul>\n<li><strong>Vulnerability scanning across 20 languages:<\/strong> Far beyond the C\/C++ scope of the benchmark <code class=\"notranslate no-translate\" data-no-translation=\"\">CyberGym<\/code>, the model achieved a vulnerability detection rate above 70% on Google's real-world source code repositories.<\/li>\n<li><strong>Automated patch generation (<code class=\"notranslate no-translate\" data-no-translation=\"\">CWE-Bench<\/code>):<\/strong> Achieved a pass@1 rate of 47.2%, comparable to the leading Frontier model (47.8%) while costing significantly less.<\/li>\n<li><strong>Chrome Security team:<\/strong> The model generated 2.6 times more accurate patches for Chrome browser vulnerabilities than the largest commercial models.<\/li>\n<li><strong>Wiz security platform:<\/strong> Increased recall by 7.5% \u2013 9.7% in internal penetration testing while reducing costs by 2.3 to 5.2 times.<\/li>\n<li><strong>Google Cloud:<\/strong> Successfully detected a critical platform vulnerability in less than 2 hours, shortening research time from months to just a few hours.<\/li>\n<\/ul><p><img decoding=\"async\" src=\"https:\/\/dps.media\/wp-content\/uploads\/2026\/09\/blog_image_9.webp\" alt=\"Real-world cybersecurity vulnerability detection rate of Gemini 3.8 Flash Cyber\" style=\"display:block; margin:20px auto; max-width:100%; height:auto;\" title=\"\"><\/p><p><img decoding=\"async\" src=\"https:\/\/dps.media\/wp-content\/uploads\/2026\/09\/blog_image_10.webp\" alt=\"Gemini 3.8 Flash Cyber optimizes vulnerability patching capabilities on CWE-Bench\" style=\"display:block; margin:20px auto; max-width:100%; height:auto;\" title=\"\"><\/p><p>The model also improves resistance to Prompt Injection attacks according to Gray Swan's evaluation, safely protecting the AI agent ecosystem when operating on servers.<\/p><h2>Gemini Flash series specification comparison table<\/h2><table style=\"width:100%; border-collapse: collapse; margin: 20px 0; font-size: 15px; text-align: left;\">\n<thead>\n<tr style=\"background-color: #151577; color: #ffffff;\">\n<th style=\"padding: 12px 14px; border: 1px solid #ddd;\">Comparison Criteria<\/th>\n<th style=\"padding: 12px 14px; border: 1px solid #ddd;\">Gemini 3.8 Flash<\/th>\n<th style=\"padding: 12px 14px; border: 1px solid #ddd;\">Gemini 3.8 Flash Cyber<\/th>\n<th style=\"padding: 12px 14px; border: 1px solid #ddd;\">Gemini 3.7 Flash<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr style=\"background-color: #f9f9f9;\">\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\"><strong>Capability focus<\/strong><\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Long-horizon coding, autonomous Agentic operation, multi-step reasoning<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Network vulnerability detection &amp; automatic source-code patch generation<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Optimized response speed and everyday operating costs<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\"><strong>Pricing (Input \/ Output)<\/strong><\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">$0.75 \/ $3.75 per million tokens<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Fairwind Program security partnership program<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">$0.75 \/ $3.75 per million tokens<\/td>\n<\/tr>\n<tr style=\"background-color: #f9f9f9;\">\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\"><strong>Key benchmark highlights<\/strong><\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Leading DeepSWE v1.1, financial &amp; legal reasoning<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">CWE-Bench patching 47.2%, bug detection &gt; 70%<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Balanced speed for high-frequency APIs<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\"><strong>Target users<\/strong><\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Software engineers, AI builders, automation-focused enterprises<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Cybersecurity experts, infrastructure operations organizations<\/td>\n<td style=\"padding: 10px 14px; border: 1px solid #ddd;\">Web application developers, conventional chatbot developers<\/td>\n<\/tr>\n<\/tbody>\n<\/table><h2>Applying Gemini 3.8 Flash to enterprise automation for SMEs<\/h2><p>For small and medium-sized enterprises (SMEs), the biggest barrier to AI adoption in the past was the high token costs of large reasoning models. With Gemini 3.8 Flash, reasonable pricing combined with robust tool-calling capabilities opens the door to process automation with tangible efficiency gains.<\/p><p>Use cases that deliver immediate value:<\/p><ul>\n<li><strong>Operating n8n automation workflows:<\/strong> Integrate the model as the central processor in n8n workflows to classify emails, extract data from multi-page invoices, and handle multichannel customer support tasks.<\/li>\n<li><strong>Security review for websites and applications:<\/strong> Automatically check syntax and scan the source code of WordPress extensions or internal software to minimize security risks before official deployment.<\/li>\n<li><strong>Moving AI prototypes into production environments:<\/strong> Applications built quickly in Google AI Studio can now be easily transformed into stable services for daily customer use.<\/li>\n<\/ul><p>If your business wants to optimize business processes and save human resources, learn more about <a href=\"https:\/\/dps.media\/en\/\">DPS.MEDIA's AI application development and process automation services<\/a> to receive a structured implementation roadmap tailored to your needs.<\/p><h2>Summary and access channels<\/h2><p>Gemini 3.8 Flash and the Cyber version are solid steps forward, moving AI from the role of a conversational assistant to that of an independent action-taking agent and source-code system protector. Persistent reasoning combined with optimized costs will provide a major boost for next-generation automation solutions.<\/p><p>You can experience the models immediately through the following platforms:<\/p><ul>\n<li><strong>Developers:<\/strong> Build agentic workflows on Google Antigravity, or connect to the API through Google AI Studio and Android Studio.<\/li>\n<li><strong>Businesses:<\/strong> Deploy at scale through the Gemini Enterprise platform.<\/li>\n<li><strong>Individual users:<\/strong> Use directly in the Gemini app, AI Mode in Google Search, and Google Sheets through the Google AI Pro or Ultra plan.<\/li>\n<\/ul><p>Need strategic consulting on applying AI to marketing, website design, or business automation? Contact DPS.MEDIA now via Hotline\/Zalo <strong>0961545445<\/strong> or visit the office at 56 Nguyen Dinh Chieu, Tan Dinh Ward, District 1, Ho Chi Minh City for detailed support from our team of experts.<\/p>","protected":false},"excerpt":{"rendered":"<p>\u0110\u00e1nh gi\u00e1 chi ti\u1ebft Gemini 3.8 Flash v\u00e0 3.8 Flash Cyber t\u1eeb Google: B\u01b0\u1edbc ti\u1ebfn \u0111\u1ed9t ph\u00e1 v\u1ec1 n\u0103ng l\u1ef1c l\u1eadp tr\u00ecnh, x\u1eed l\u00fd \u0111a b\u01b0\u1edbc v\u00e0 t\u1ef1 \u0111\u1ed9ng v\u00e1 l\u1ed7i b\u1ea3o m\u1eadt m\u1ea1ng.<\/p>","protected":false},"author":0,"featured_media":40556,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1799,1635,2410],"tags":[675],"class_list":["post-40561","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-digital-marketing","category-tin-tuc-cong-nghe","tag-dps-media"],"acf":[],"rankmath_keywords":{"primary":"Gemini 3.8 Flash, Gemini 3.8 Flash Cyber, Google Gemini 3.8, \u0111\u00e1nh gi\u00e1 Gemini 3.8 Flash, t\u00ednh n\u0103ng Gemini 3.8 Flash","secondary":["Gemini 3.8 Flash Cyber","Google Gemini 3.8","\u0111\u00e1nh gi\u00e1 Gemini 3.8 Flash","t\u00ednh n\u0103ng Gemini 3.8 Flash"]},"yoast_keywords":{"primary":"","secondary":[]},"yoast_focuskw":"","rankmath_focuskw":"Gemini 3.8 Flash, Gemini 3.8 Flash Cyber, Google Gemini 3.8, \u0111\u00e1nh gi\u00e1 Gemini 3.8 Flash, t\u00ednh n\u0103ng Gemini 3.8 Flash","seo_keywords":{"primary":"Gemini 3.8 Flash, Gemini 3.8 Flash Cyber, Google Gemini 3.8, \u0111\u00e1nh gi\u00e1 Gemini 3.8 Flash, t\u00ednh n\u0103ng Gemini 3.8 Flash","secondary":["Gemini 3.8 Flash Cyber","Google Gemini 3.8","\u0111\u00e1nh gi\u00e1 Gemini 3.8 Flash","t\u00ednh n\u0103ng Gemini 3.8 Flash"]},"_links":{"self":[{"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/posts\/40561","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/comments?post=40561"}],"version-history":[{"count":0,"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/posts\/40561\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/media\/40556"}],"wp:attachment":[{"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/media?parent=40561"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/categories?post=40561"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dps.media\/en\/wp-json\/wp\/v2\/tags?post=40561"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}