{
  "id": 12646406,
  "title": "Mistral Large 4 Takes on GPT-6 Astra With 1 Trillion Parameters Model",
  "url": "https://urgent.news/2026/10/07/mistral-large-4-takes-on-gpt-6-astra-with-1-trillion-parameters-model",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-07T14:35:21.000Z",
  "source": {
    "name": "ProPakistani",
    "slug": "propakistani",
    "url": "https://propakistani.pk/2026/10/07/mistral-large-4-takes-on-gpt-6-astra-with-1-trillion-parameters-model/"
  },
  "original_language": "en",
  "account": "Mistral AI has made its latest large language model, Mistral Large 4, available in public preview. This model boasts a staggering 1 trillion parameters organized using a mixture-of-experts architecture, though only 49 billion parameters are activated at any given time. This design choice improves hardware efficiency by not utilizing the full parameter set for every request. The model can answer questions in over 160 languages.\n\nMistral Large 4 has demonstrated strong performance on various benchmarks. It earned a top-five score on the AA Cyber Index, indicating its ability to identify and fix software vulnerabilities. The model's coding capabilities are noteworthy, as it outperformed open-source rivals such as Qwen 3.8 Max and DeepSeek V4 Pro. Additionally, it scored higher than DeepSeek V4 Pro on AutomationBench and AA-Briefcase, which involve tasks that could take humans weeks to accomplish.\n\nMistral trained Large 4 using 3,800 Nvidia Grace Blackwell chips, which are optimized for AI workloads. Each chip combines two graphics cards with one CPU. Mistral's training workflow generated around 33 billion tokens per day, with approximately 33 billion tokens used in refining Large 4. The model's development also involved a parallel rollout system capable of running tens of thousands of AI rollouts simultaneously. This infrastructure will be used to create even larger and more capable versions of the model in the future.",
  "summary": "Mistral AI has opened access to Mistral Large 4, its most capable large language model to date. The model is … Read More The post Mistral Large 4 Takes on GPT-6 Astra With 1 Trillion Parameters Model appeared first on ProPakistani .",
  "key_points": [],
  "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."
}