{
  "id": 5122354,
  "title": "Introducing Gemini 3.8 Flash and 3.8 Flash Cyber",
  "url": "https://urgent.news/2026/09/02/introducing-gemini-3-8-flash-and-3-8-flash-cyber-5122354",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T16:18:31.000Z",
  "source": {
    "name": "Google DeepMind",
    "slug": "google-deepmind",
    "url": "https://deepmind.google/blog/introducing-gemini-3-8-flash-and-38-flash-cyber/"
  },
  "original_language": "en",
  "account": "Google has unveiled Gemini 3.8, a cutting-edge reasoning and coding model, accompanied by its Flash variant, Gemini 3.8 Flash and 3.8 Flash Cyber. This release is the third in six weeks, building upon the success of the 3.7 Flash model. Gemini 3.8 introduces two variants, both powered by the same foundational intelligence, and enhanced by long-running agentic loops to recursively evaluate and refine the models. These improvements were achieved through rigorous training in cybersecurity, resulting in significant gains in coding and reasoning abilities.\n\nGemini 3.8 Flash delivers performance nearly comparable to higher-cost frontier models, often matching or surpassing larger models on DeepSWE v1.1, a long-horizon software engineering benchmark. It also outperforms many larger models in various quantitative and professional fields, including finance, legal analysis, and multi-step reasoning across diverse domains. This model operates with greater diligence on complex tasks, executing additional reasoning steps and calling tools iteratively. For applications with compute efficiency as the primary constraint, developers can utilize lower effort levels to minimize token overhead or continue using Gemini 3.7 Flash.\n\nFor those in the cybersecurity domain, Gemini 3.8 Flash Cyber offers a distinct advantage. Available to a select group of trusted defenders through the Fairwind Program, this variant is powered by Flash speed and cost, enabling quick iteration. On the CyberGym benchmark, Gemini 3.8 Flash Cyber demonstrates frontier-level performance in autonomous vulnerability discovery, surpassing both 3.5 Flash Cyber and significantly larger frontier models. In a comprehensive internal benchmark evaluating the model's capability to discover vulnerabilities across complex codebases spanning 20 programming languages, Gemini 3.8 Flash Cyber achieved a success rate exceeding 70%.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Google Blog",
        "title": "Introducing Gemini 3.8 Flash and 3.8 Flash Cyber",
        "url": "https://urgent.news/2026/09/02/introducing-gemini-3-8-flash-and-3-8-flash-cyber",
        "published": "2026-09-02T15:00:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}