{
  "id": 14007192,
  "title": "Microsoft’s New AI Model Is 35x Faster and Up to 200x Cheaper Than GPT-6 Sol",
  "url": "https://urgent.news/2026/10/12/microsofts-new-ai-model-is-35x-faster-and-up-to-200x-cheaper-than-gpt",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-12T16:44:58.000Z",
  "source": {
    "name": "ProPakistani",
    "slug": "propakistani",
    "url": "https://propakistani.pk/2026/10/12/microsofts-new-ai-model-is-35x-faster-and-up-to-200x-cheaper-than-gpt-6-sol/"
  },
  "original_language": "en",
  "account": "Microsoft has introduced Microsoft-Decision-1, a compact AI model tailored for decision-making and classification tasks, as opposed to generating lengthy text responses. Based on Alibaba's open-weight Qwen3.5-9B, Microsoft has fine-tuned Decision-1 for specific tasks like classification, routing, verification, and workflow control. The model demonstrates superior performance and efficiency compared to much larger general-purpose AI models, while also consuming significantly less computational resources.\n\nKey characteristics of Decision-1 include receiving a fixed set of possible answers and assigning calibrated probability scores to each option. It can manage yes/no inquiries, multiple-choice tasks, and ratings. Microsoft has designed Decision-1 to assess AI responses or agent actions against predefined criteria. Potential applications for this model span various domains, such as classification, routing, prioritization, verification, workflow control, AI response grading, and agent action evaluation.\n\nUnlike conventional chatbots, Decision-1 is not designed for composing essays, emails, or conversational answers. Microsoft emphasizes the model's specialization in decision-making and classification, rather than generating human-like text. Presently, Decision-1 is priced at $0.042 per million input tokens, with output tokens being free of charge. The pricing structure aligns with the model's purpose, as it primarily produces scores for a predefined set of choices, rather than extensive text output.\n\nIn Microsoft's internal tests, Decision-1 outperformed other models in 36 benchmark challenges involving nearly 150,000 questions. The evaluation revealed that Decision-1 achieved the highest overall accuracy among the models presented, while also showcasing remarkable speed. Specifically, Decision-1 was found to be 2.5 times faster than H2O-Lightning-4B v1.1 and an impressive 35 times faster than GPT-6 Sol.\n\nMicrosoft has also assessed Decision-1's consistency when presented with identical questions rephrased or ordered differently. The model exhibited only a 1.3% decision change on average, which is crucial for tasks like routing and automated workflows, where minor alterations in phrasing should not yield drastically different decisions. In practical applications, such as Xbox Research's categorization of over 10,000 pieces of player feedback and reviews, Decision-1 demonstrated performance competitive with GPT-6 Sol while being 14 times faster and approximately 200 times cheaper.\n\nMicrosoft plans to continue refining Decision-1 by rebasing it on other models, potentially including OpenAI models and Microsoft's own MAI models. The company acknowledges the growing significance of decision models within the AI landscape, with various competitors like TypeSafe AI, OpenAI, Cloudflare, Strands, Liquid AI, Perplexity, and Snowflake also developing decision-focused AI systems.",
  "summary": "Microsoft has launched Microsoft-Decision-1, a small AI model designed to make decisions between predefined options instead of generating long-form responses. … Read More The post Microsoft’s New AI Model Is 35x Faster and Up to 200x Cheaper Than GPT-6 Sol 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."
}