{
  "id": 20475374,
  "title": "MAI Code 2 and Code 2 Flash",
  "url": "https://urgent.news/2026/11/17/mai-code-2-and-code-2-flash",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-17T18:25:13.000Z",
  "source": {
    "name": "Hacker News",
    "slug": "hacker-news",
    "url": "https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/"
  },
  "original_language": "en",
  "account": "MAI Code 2 and Code 2 Flash are two new coding models designed for everyday enterprise coding tasks. MAI Code 2 is a top performer in terms of price-performance, excelling at tasks such as code production, deep debugging, multi-file implementation, repo-wide reasoning, and UI work. It also produces code changes that maintainers would merge, at a fraction of the cost of competing models. In contrast, MAI Code 2 Flash is optimized for high-volume work, including code explanations, targeted edits, routine fixes, and rapid iterations. Both models prioritize the coding work done daily by customers and are not yet tuned for more challenging problems like fundamental science research or very long horizon agentic work.\n\nMAI Code 2 outperforms GPT-6.1 Sol by 98% on average across four widely-used benchmarks that focus on common enterprise coding tasks. It also matches 96% of Claude Opus 5.5's performance while costing just 5% as much. The model is particularly adept at creating dashboards and visualizations, outperforming GPT-6 Sol in blind frontend app evaluations 67% of the time. MAI Code 2 Flash offers comparable performance to GPT-6 Luna and Claude Haiku 5.5 at just 65% and 20% of the cost, respectively, across three coding benchmarks measuring day-to-day, high-volume routine coding tasks.\n\nThe model is becoming a key component of Microsoft products, powering features in Excel, PowerPoint, Word, Teams, and Fabric IQ. In Excel, MAI Code 2 performs on par with GPT-6 Luna in terms of accuracy and token efficiency. In PowerPoint, it matches GPT-6 Luna's quality and token efficiency in key production benchmark scenarios. In Word, the model performs similarly to GPT-6.1 Sol on five key internal benchmarks while using 42% fewer input tokens per call. For Teams, MAI Code 2 beats GPT-6 Luna in tracking decisions, filling knowledge gaps, and adapting custom instructions, all while using up to 28% fewer tokens for instruction following and tool calling. In offline evaluations, the model also outperforms GPT-5.6-sol, GPT-6-astra, and Opus-5.5 at DAX query generation, a critical capability for Power BI semantic models.\n\nMAI Code 2 and Code 2 Flash are currently available in GitHub Copilot, with Code 2 Flash in limited-access preview in Microsoft Foundry. They are set to be released on Open Router, Vercel, Hermes, Open Code, and Kilo in the near future. Custom models can be further tailored on the Custom Models page. These results represent evaluations on specific coding tasks and configurations, and performance and cost comparisons may vary based on workload, model configuration, reasoning effort, and pricing assumptions.",
  "summary": null,
  "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."
}