{
  "id": 5123229,
  "title": "Gemini 3.8 Flash and 3.8 Flash Cyber",
  "url": "https://urgent.news/2026/09/02/gemini-3-8-flash-and-3-8-flash-cyber",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T15:44:06.000Z",
  "source": {
    "name": "Hacker News",
    "slug": "hacker-news",
    "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/"
  },
  "original_language": "en",
  "account": "The Gemini 3.8 model, unveiled today, represents a significant leap forward in reasoning and coding capabilities, following the highly successful 3.7 Flash release three weeks prior. This latest iteration is available in two variants, both powered by the same foundational intelligence, yet optimized for different deployment environments. The innovations driving the enhancements include rigorous training in cybersecurity and the introduction of long-running agentic loops to recursively evaluate and refine the underlying models.\n\nGemini 3.8 Flash delivers substantial improvements over its predecessor, often matching the performance of higher-cost frontier models. This variant excels in complex engineering tasks, outperforming most larger models in end-to-end solutions, all while being more cost-effective. It also demonstrates exceptional dependability for critical enterprise applications across various specialized knowledge domains.\n\nIn quantitative and professional fields requiring advanced analysis, Gemini 3.8 Flash outperforms both its predecessor and other frontier models in benchmarks such as Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. It achieves a notable 54.9% score on HLE-Verified, showcasing its ability to handle multi-step reasoning across diverse fields like STEM, humanities, and professional areas.\n\nFor applications where compute efficiency is paramount, developers can utilize Gemini 3.8 Flash's lower effort levels to minimize token overhead, though the model naturally opts to execute more reasoning steps and use more tokens to achieve optimal performance on complex tasks. For users in the trusted \"Fairwind Program,\" Gemini 3.8 Flash Cyber offers a unique advantage in cybersecurity, boasting frontier-level performance in autonomous vulnerability discovery on industry-standard benchmarks like CyberGym.\n\nThe model significantly surpasses both the 3.5 Flash Cyber version and larger frontier models, achieving an impressive success rate of over 70% across a wide range of vulnerabilities in complex codebases spanning 20 programming languages. Gemini 3.8 Flash Cyber is specifically designed to equip defenders with expert capabilities, prioritizing vulnerability fixing over offensive exploits, as per the Frontier Safety Framework. This model is already being used to enhance code security at Google, where it is deployed with safeguards against misuse in CBRN and cyber offense domains.\n\nGemini 3.8 models also demonstrate remarkable improvements in prompt injection robustness, as measured by Gray Swan, making them highly resilient against malicious attacks targeting prompt injection. The introductory pricing for Gemini 3.8 Flash Cyber expires on December 31, 2026, following which a new pricing model will apply, starting at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens from January 1, 2027.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Hacker News Best",
        "title": "Gemini 3.8 Flash and 3.8 Flash Cyber",
        "url": "https://urgent.news/2026/09/02/gemini-3-8-flash-and-3-8-flash-cyber-5130652",
        "published": "2026-09-02T15:12:40.000Z"
      }
    ]
  },
  "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."
}