{
  "id": 8980593,
  "title": "Grok 4.7",
  "url": "https://urgent.news/2026/09/21/grok-4-7-8980593",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-21T15:50:15.000Z",
  "source": {
    "name": "Hacker News Best",
    "slug": "hacker-news-best",
    "url": "https://x.ai/news/grok-4-7"
  },
  "original_language": "en",
  "account": "SpaceXAI has unveiled Grok 4.7, its most powerful model for coding and knowledge work, which is twice as fast and half the price of comparable models. This new model is designed to work longer on challenging tasks, verify its own work more diligently, and come with the best-calibrated safeguards yet. Released at the same price and speed as Grok 4.6, Grok 4.7 holds its own in the class, particularly on CursorBench 4.0, which focuses on extended coding tasks.\n\nPowered by a larger base model, Grok 4.7 underwent a more extensive reinforcement learning process on a tougher mix of tasks, with an emphasis on problems that require several hours to solve. The model excels at self-assessment and managing larger context, while also understanding the Grok Bot harness natively, enhancing its conversational abilities and general knowledge work performance.\n\nIn practical applications, Grok 4.7 outperforms its predecessor in GDPval and AA Briefcase benchmarks, where it's asked to perform tasks typically handled by professionals like lawyers, nurses, and financial analysts. The model's safety features are also top-notch, especially in dual-use domains such as cybersecurity and biological work. It leads in utility for benign tasks and safe refusal in dangerous ones, surpassing LatchBio's biosafety benchmark at 62.4%.\n\nGrok 4.7 strikes a balance between strong cyber defense capabilities and low refusal rates for legitimate use, allowing only 3.3% of risky, malicious cyber prompts through while seldom blocking legitimate security work. The model has started providing select cybersecurity partners with invite-only access to its red-team capabilities for defense research.\n\nGrok 4.7 is now accessible in Cursor, Grok Build, and via the Grok API, third-party coding harnesses, model routers, and cloud platforms. Pricing begins at $2 per million input tokens and $6 per million output tokens, with a faster variant available at twice the output speed for twice the price.",
  "summary": "Article URL: https://x.ai/news/grok-4-7 Comments URL: https://news.ycombinator.com/item?id=49788838 Points: 281 # Comments: 223",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Hacker News",
        "title": "Grok 4.7",
        "url": "https://urgent.news/2026/09/21/grok-4-7",
        "published": "2026-09-21T15:50:15.000Z"
      }
    ]
  },
  "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."
}