{
  "version": 1,
  "event_id": "evt_d5ced273c2ef0a66",
  "url": "https://xiyu.news/events/evt_d5ced273c2ef0a66/",
  "json": "https://xiyu.news/api/events/evt_d5ced273c2ef0a66.json",
  "type": "product_release",
  "status": "monitoring",
  "category": "technology",
  "title": {
    "zh": "Gemini 4 Argon：我们通往前沿智能的下一个时代",
    "en": "Gemini 4 Argon: our next era of frontier intelligence"
  },
  "current_state": {
    "zh": "谷歌发布了 Gemini 4 Argon，并将其作为新一代旗舰长程模型。单次输出上限从此前 64K tokens 提高到 100 万 tokens。模型暂不向普通用户开放，首批仅通过 Fairwind Program 提供给可信网络安全团队，后续再逐步开放给付费 API 客户和 Google AI Ultra 用户。API 定价初期为每百万输入 tokens 2 美元、输出 10 美元，之后翻倍至 4 美元和 20 美元。官方评测中，Argon 在 DeepSWE v1.1 得 77.9%，高于 GPT-6 Astra 和 Claude Opus 5.5，但在 FrontierSWE、Terminal-Bench 4.0 及部分科学任务上仍落后。谷歌重点强化了网络安全能力，可信防御团队版本甚至会取消部分网络安全限制；正式开放前，谷歌仍在测试滥用、提示词注入和 Agent 越界等风险。",
    "en": "Google released Gemini 4 Argon as its next flagship long-horizon model. Its single-pass output limit jumps from 64K tokens to 1 million tokens. The model is not yet open to ordinary users; initial access is only through the Fairwind Program to trusted cybersecurity teams, with later rollout to paid API customers and Google AI Ultra users. API pricing starts at $2 per million input tokens and $10 per million output tokens, later doubling to $4 and $20. In official evaluations, Argon scores 77.9% on DeepSWE v1.1, ahead of GPT-6 Astra and Claude Opus 5.5, but still trails on FrontierSWE, Terminal-Bench 4.0 and some scientific tasks. Google emphasized cybersecurity capabilities, with trusted defender versions even removing some cyber restrictions; before formal opening, Google is still testing abuse, prompt injection and agent boundary risks."
  },
  "first_seen_at": "2026-09-30T23:26:56.671688+00:00",
  "last_updated_at": "2026-10-01T00:03:18.016338+00:00",
  "last_material_change_at": "2026-10-01T00:03:18.016338+00:00",
  "confidence": 0.75,
  "updates_count": 2,
  "sources_count": 2,
  "entities": [
    "argon",
    "gemini",
    "gemini-4-argon",
    "token"
  ],
  "identifiers": [],
  "topics": [
    "ai-labs",
    "ai-model-release",
    "ai-safety",
    "frontier-models",
    "gemini-4-argon",
    "google-deepmind",
    "koray-kavukcuoglu",
    "long-context"
  ],
  "updates": [
    {
      "update_id": "upd_2255c23391117659",
      "event_id": "evt_d5ced273c2ef0a66",
      "occurred_at": "2026-09-30T20:01:45Z",
      "published_at": "2026-09-30T20:01:45Z",
      "first_seen_at": "2026-09-30T23:26:56.671688Z",
      "time_precision": "published",
      "update_type": "initial",
      "material_change": true,
      "title_zh": "Gemini 4 Argon：我们通往前沿智能的下一个时代",
      "title_en": "Gemini 4 Argon: our next era of frontier intelligence",
      "what_changed_zh": "谷歌发布了其新一代前沿 AI 模型 Gemini 4 Argon。谷歌首席 AI 架构师、Google DeepMind 高级副总裁 Koray Kavukcuoglu 表示，该模型在“真实软件工程、法律与金融等企业知识工作，以及网络安全防御等复杂工作流中提供前沿性能”。谷歌目前限制该模型的访问权限，称将继续从早期测试者处收集反馈并迭代护栏机制，之后再向开发者、企业和消费者开放 Argon。\n\n该发布使顶级实验室之间的竞争再添一款旗舰模型，但访问权限仍受限——开发者、企业和消费者目前尚不能普遍使用，谷歌给出的理由是仍在完善护栏机制并收集早期测试者反馈。\n\n谷歌称 Argon 在编程、推理和多模态方面具备能力，并能在企业工作流中持续完成长链条、多步骤任务。Hacker News 评论者引用公告称，Argon 智能体正在谷歌内部将 C/C++ 代码库迁移至 Rust；独立评测机构 Artificial Analysis 将 Gemini 4 Argon（High）列为智能水平领先的模型之一，且相对于同等价位的模型定价合理。",
      "what_changed_en": "Google unveiled Gemini 4 Argon, its next frontier AI model, which the company says delivers \"frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense,\" according to chief AI architect and Google DeepMind SVP Koray Kavukcuoglu. Google is limiting access to the model for now, saying it will keep gathering feedback from early testers as it iterates on guardrails before making Argon available to developers, enterprises and consumers.\n\nThe announcement puts a new flagship model into the top-lab competition, but access remains restricted — developers, enterprises and consumers cannot yet use it generally, with Google citing continued work on guardrails and feedback from early testers as the reason.\n\nGoogle describes Argon's capabilities across coding, reasoning and multimodality, and its ability to sustain long, multi-step tasks in enterprise workflows. Hacker News commenters quoted the announcement as saying Argon agents are working on migrating C/C++ codebases to Rust across Google; independent evaluator Artificial Analysis rates Gemini 4 Argon (High) as among the leading models in intelligence and reasonably priced relative to models at a similar price point.",
      "current_state_zh": "谷歌发布了其新一代前沿 AI 模型 Gemini 4 Argon。谷歌首席 AI 架构师、Google DeepMind 高级副总裁 Koray Kavukcuoglu 表示，该模型在“真实软件工程、法律与金融等企业知识工作，以及网络安全防御等复杂工作流中提供前沿性能”。谷歌目前限制该模型的访问权限，称将继续从早期测试者处收集反馈并迭代护栏机制，之后再向开发者、企业和消费者开放 Argon。\n\n该发布使顶级实验室之间的竞争再添一款旗舰模型，但访问权限仍受限——开发者、企业和消费者目前尚不能普遍使用，谷歌给出的理由是仍在完善护栏机制并收集早期测试者反馈。\n\n谷歌称 Argon 在编程、推理和多模态方面具备能力，并能在企业工作流中持续完成长链条、多步骤任务。Hacker News 评论者引用公告称，Argon 智能体正在谷歌内部将 C/C++ 代码库迁移至 Rust；独立评测机构 Artificial Analysis 将 Gemini 4 Argon（High）列为智能水平领先的模型之一，且相对于同等价位的模型定价合理。",
      "current_state_en": "Google unveiled Gemini 4 Argon, its next frontier AI model, which the company says delivers \"frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense,\" according to chief AI architect and Google DeepMind SVP Koray Kavukcuoglu. Google is limiting access to the model for now, saying it will keep gathering feedback from early testers as it iterates on guardrails before making Argon available to developers, enterprises and consumers.\n\nThe announcement puts a new flagship model into the top-lab competition, but access remains restricted — developers, enterprises and consumers cannot yet use it generally, with Google citing continued work on guardrails and feedback from early testers as the reason.\n\nGoogle describes Argon's capabilities across coding, reasoning and multimodality, and its ability to sustain long, multi-step tasks in enterprise workflows. Hacker News commenters quoted the announcement as saying Argon agents are working on migrating C/C++ codebases to Rust across Google; independent evaluator Artificial Analysis rates Gemini 4 Argon (High) as among the leading models in intelligence and reasonably priced relative to models at a similar price point.",
      "detailed_summary_zh": "谷歌发布了其新一代前沿 AI 模型 Gemini 4 Argon。谷歌首席 AI 架构师、Google DeepMind 高级副总裁 Koray Kavukcuoglu 表示，该模型在“真实软件工程、法律与金融等企业知识工作，以及网络安全防御等复杂工作流中提供前沿性能”。谷歌目前限制该模型的访问权限，称将继续从早期测试者处收集反馈并迭代护栏机制，之后再向开发者、企业和消费者开放 Argon。\n\n该发布使顶级实验室之间的竞争再添一款旗舰模型，但访问权限仍受限——开发者、企业和消费者目前尚不能普遍使用，谷歌给出的理由是仍在完善护栏机制并收集早期测试者反馈。\n\n谷歌称 Argon 在编程、推理和多模态方面具备能力，并能在企业工作流中持续完成长链条、多步骤任务。Hacker News 评论者引用公告称，Argon 智能体正在谷歌内部将 C/C++ 代码库迁移至 Rust；独立评测机构 Artificial Analysis 将 Gemini 4 Argon（High）列为智能水平领先的模型之一，且相对于同等价位的模型定价合理。",
      "detailed_summary_en": "Google unveiled Gemini 4 Argon, its next frontier AI model, which the company says delivers \"frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense,\" according to chief AI architect and Google DeepMind SVP Koray Kavukcuoglu. Google is limiting access to the model for now, saying it will keep gathering feedback from early testers as it iterates on guardrails before making Argon available to developers, enterprises and consumers.\n\nThe announcement puts a new flagship model into the top-lab competition, but access remains restricted — developers, enterprises and consumers cannot yet use it generally, with Google citing continued work on guardrails and feedback from early testers as the reason.\n\nGoogle describes Argon's capabilities across coding, reasoning and multimodality, and its ability to sustain long, multi-step tasks in enterprise workflows. Hacker News commenters quoted the announcement as saying Argon agents are working on migrating C/C++ codebases to Rust across Google; independent evaluator Artificial Analysis rates Gemini 4 Argon (High) as among the leading models in intelligence and reasonably priced relative to models at a similar price point.",
      "background_zh": "Argon 是谷歌 Gemini 旗舰系列的后续产品：Gemini 3 于 11 月发布，而 DeepMind 首席技术官 Koray Kavukcuoglu 担任新设立的谷歌首席 AI 架构师一职，职责是加快公司 AI 技术的产品化。“前沿模型”是产业与政策领域的术语，指处于当前通用人工智能能力最前沿或接近最前沿的模型，而非固定的技术类别。",
      "background_en": "Argon arrives as the successor in Google's Gemini flagship line: Gemini 3 was released in November, and Koray Kavukcuoglu, DeepMind's chief technology officer, holds the newly created role of Google chief AI architect aimed at accelerating productization of the company's AI technology. \"Frontier model\" is an industry and policy term for a model at or near the current leading edge of general-purpose AI capability, rather than a fixed technical class.",
      "community_discussion_zh": "Hacker News 上的评论意见不一。一些人对受限的发布方式提出批评，指出非 Flash 模型再次对普通订阅用户无限期不可用，并质疑付费档位的价值；也有人认为各实验室之间的持续交替领先，削弱了“AI 是赢者通吃、先发优势永不丢失”这一与 Anthropic 的 Dario Amodei 相关的看法。另有用户描述了一次 Gemini 3.8 Flash 的使用经历：该模型将 GDB 附加到 GPU 驱动、逆向内核队列 ioctl 接口，并编写 LD_PRELOAD 垫片，使 ROCm 版 llama.cpp 在 Strix Halo 机器上运行起来。",
      "community_discussion_en": "Commenters on Hacker News were divided. Several criticized the restricted rollout, noting that a non-Flash model is again unavailable to regular subscribers for an indefinite period and questioning the value of paid tiers, while one argued the steady leapfrogging between labs undercuts the idea — associated with Anthropic's Dario Amodei — that AI is a winner-takes-all field where an early lead is never ceded. Another user described a Gemini 3.8 Flash session in which the model attached GDB to a GPU driver, reverse-engineered a kernel queue ioctl interface and wrote an LD_PRELOAD shim to get ROCm llama.cpp working on a Strix Halo machine.",
      "market_impact_zh": "与加密市场相关的传导渠道主要是情绪面，而非公告披露的任何集成：大型实验室发布前沿模型往往会推动围绕 AI 相关代币和去中心化算力代币的叙事，尽管该公告本身并未披露与加密市场在托管、流动性、上币或监管方面的任何关联。任何影响都将通过交易者在这类 AI 主题板块上的仓位体现，而非通过目前仍受限的 Gemini 4 Argon 本身的可获得性。",
      "market_impact_en": "The crypto-relevant transmission channel is sentiment rather than any disclosed integration: frontier-model releases from major labs tend to feed narratives around AI-adjacent and decentralized compute tokens, even though the announcement itself discloses no custody, liquidity, listing or regulatory linkage to crypto markets. Any effect would run through trader positioning on those AI-themed segments rather than through Gemini 4 Argon's own availability, which remains restricted.",
      "importance_score": 8.5,
      "references": [
        {
          "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
          "title": "Introducing Gemini 4 Argon"
        },
        {
          "url": "https://artificialanalysis.ai/models/gemini-4-argon",
          "title": "Gemini 4 Argon (high) - Intelligence, Performance... | Artificial Analysis"
        },
        {
          "url": "https://www.ft.com/content/3b477836-8a87-4ce3-bea1-719f28952a9a?syn-25a6b1a6=1",
          "title": "Google’s Koray Kavukcuoglu: Turning abstract AI thinking into..."
        }
      ],
      "confidence": 0.75,
      "story_ids": [
        "rss:deepmind.google_blog_rss.xml:ce56aa83d0064ef1"
      ],
      "sources": [
        {
          "url": "https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/",
          "label": "Google DeepMind Blog",
          "source_type": "rss",
          "official": false
        }
      ]
    },
    {
      "update_id": "upd_76eb1e1aea0707d2",
      "event_id": "evt_d5ced273c2ef0a66",
      "occurred_at": "2026-09-30T23:50:02Z",
      "published_at": "2026-09-30T23:50:02Z",
      "first_seen_at": "2026-10-01T00:03:18.016338Z",
      "time_precision": "published",
      "update_type": "confirmation",
      "material_change": true,
      "title_zh": "Gemini 4 Argon发布：100万Token输出，首批只给可信安全团队",
      "title_en": "Gemini 4 Argon发布：100万Token输出，首批只给可信安全团队",
      "what_changed_zh": "新报道补充了 Gemini 4 Argon 的关键细节：单次输出上限从 64K tokens 提升至 100 万 tokens；首批仅通过 Fairwind Program 向可信网络安全团队开放，后续再逐步开放给付费 API 客户和 Google AI Ultra 用户；API 上线初期按每百万输入 tokens 2 美元、输出 10 美元收费，之后价格翻倍至 4 美元和 20 美元；在 DeepSWE v1.1 得 77.9%，领先 GPT-6 Astra 的 74.1% 和 Claude Opus 5.5 的 74.2%，但在 FrontierSWE、Terminal-Bench 4.0 和部分科学任务上仍落后；网络安全能力被重点强化，可信防御团队版本会取消部分网络安全限制，同时谷歌仍在测试滥用、提示词注入和 Agent 越界等风险。",
      "what_changed_en": "The new report adds key details about Gemini 4 Argon: its single-pass output limit jumps from 64K tokens to 1 million tokens; initial access is only through the Fairwind Program to trusted cybersecurity teams, with later rollout to paid API customers and Google AI Ultra users; API pricing starts at $2 per million input tokens and $10 per million output tokens, later doubling to $4 and $20; it scores 77.9% on DeepSWE v1.1, ahead of GPT-6 Astra's 74.1% and Claude Opus 5.5's 74.2%, but still trails on FrontierSWE, Terminal-Bench 4.0 and some scientific tasks; cybersecurity capabilities are emphasized, with trusted defender versions removing some cyber restrictions, while Google continues testing abuse, prompt injection and agent boundary risks.",
      "current_state_zh": "谷歌发布了 Gemini 4 Argon，并将其作为新一代旗舰长程模型。单次输出上限从此前 64K tokens 提高到 100 万 tokens。模型暂不向普通用户开放，首批仅通过 Fairwind Program 提供给可信网络安全团队，后续再逐步开放给付费 API 客户和 Google AI Ultra 用户。API 定价初期为每百万输入 tokens 2 美元、输出 10 美元，之后翻倍至 4 美元和 20 美元。官方评测中，Argon 在 DeepSWE v1.1 得 77.9%，高于 GPT-6 Astra 和 Claude Opus 5.5，但在 FrontierSWE、Terminal-Bench 4.0 及部分科学任务上仍落后。谷歌重点强化了网络安全能力，可信防御团队版本甚至会取消部分网络安全限制；正式开放前，谷歌仍在测试滥用、提示词注入和 Agent 越界等风险。",
      "current_state_en": "Google released Gemini 4 Argon as its next flagship long-horizon model. Its single-pass output limit jumps from 64K tokens to 1 million tokens. The model is not yet open to ordinary users; initial access is only through the Fairwind Program to trusted cybersecurity teams, with later rollout to paid API customers and Google AI Ultra users. API pricing starts at $2 per million input tokens and $10 per million output tokens, later doubling to $4 and $20. In official evaluations, Argon scores 77.9% on DeepSWE v1.1, ahead of GPT-6 Astra and Claude Opus 5.5, but still trails on FrontierSWE, Terminal-Bench 4.0 and some scientific tasks. Google emphasized cybersecurity capabilities, with trusted defender versions even removing some cyber restrictions; before formal opening, Google is still testing abuse, prompt injection and agent boundary risks.",
      "detailed_summary_zh": "Google DeepMind released Gemini 4 Argon, a flagship long-horizon model whose single-pass output limit jumps from 64K to 1 million tokens, initially offered only to trusted cybersecurity teams via the Fairwind Program before wider API and consumer rollout.",
      "detailed_summary_en": "Google DeepMind released Gemini 4 Argon, a flagship long-horizon model whose single-pass output limit jumps from 64K to 1 million tokens, initially offered only to trusted cybersecurity teams via the Fairwind Program before wider API and consumer rollout.",
      "background_zh": "",
      "background_en": "",
      "community_discussion_zh": "",
      "community_discussion_en": "",
      "market_impact_zh": "",
      "market_impact_en": "",
      "importance_score": 8.5,
      "references": [],
      "confidence": 0.95,
      "story_ids": [
        "telegram:theblockbeats:198508"
      ],
      "sources": [
        {
          "url": "https://m.theblockbeats.info/flash/369826?from=telegram",
          "label": "theblockbeats",
          "source_type": "telegram",
          "official": false
        }
      ]
    }
  ]
}