{
  "id": 8952427,
  "title": "SpaceXAI releases Grok 4.7, which it says is better at verifying its own work and managing longer context, available for $2/1M input and $6/1M output tokens (xAI)",
  "url": "https://urgent.news/2026/09/21/spacexai-releases-grok-4-7-which-it-says-is-better-at-verifying-its",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-21T15:55:01.000Z",
  "source": {
    "name": "Techmeme",
    "slug": "techmeme",
    "url": "https://x.ai/news/grok-4-7"
  },
  "original_language": "en",
  "account": "SpaceXAI has introduced Grok 4.7, their latest and most advanced model for coding and knowledge work. This new model demonstrates twice the speed and half the cost of its competitors, while also offering enhanced capabilities. Grok 4.7 shines in longer, more challenging tasks, as it retains a higher level of accuracy in self-checking and better manages extensive context. It comes at the same price and speed as its predecessor, Grok 4.6, making it highly competitive in its class.\n\nThe model boasts a larger base compared to its predecessor, Grok 4.6, and underwent a more extensive reinforcement learning training process on a tougher mix of tasks, specifically tailored for problems that require hours to complete. Grok 4.7 excels in verifying its own work and handling extended context, and it has been fine-tuned to understand the Grok Bot harness, leading to improvements in conversational tasks and general knowledge work.\n\nGrok 4.7 demonstrates its prowess in creating documents and presentations, performing comparably to other frontier models in GDPval and AA Briefcase benchmarks, which assess professionals like lawyers, nurses, and financial analysts. It outperforms Grok 4.6 on both benchmarks and maintains its position among the leading frontier models. Built with an entirely new safeguard stack, Grok 4.7 is the strongest model tested for refusal and jailbreak resistance.\n\nIn dual-use domains such as cybersecurity and biological work, Grok 4.7 excels in utility for benign tasks and safe refusal on dangerous ones. It tops LatchBio’s biosafety benchmark at 62.4%, striking a balance between strong cyber defense capabilities and low refusal rates for legitimate use. The model shows the highest safety on HackerBench v0.3, our benchmark for risky and malicious cyber tasks, permitting only 3.3% of risky dual-use prompts while rarely blocking legitimate security work.\n\nSpaceXAI has also initiated invite-only access to Grok 4.7’s red-team capabilities for defense research, granted to select cybersecurity partners. The model is now available in Cursor and Grok Build, as well as through the Grok API, third-party coding harnesses, and model routers and cloud platforms. Pricing starts at $2 per million input tokens and $6 per million output tokens, with a fast variant offering double the output speed at double the price.",
  "summary": "SpaceXAI's most powerful model for coding and knowledge work. Twice as fast, at half the price of comparable models.",
  "key_points": [
    "Grok 4.7 is SpaceXAI's latest AI model for coding and knowledge work",
    "Model offers twice the speed and half the cost of competitors",
    "Grok 4.7 excels in self-checking and handling extensive context"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "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."
}