{ "version": 1, "event_id": "evt_48dd263d7b649072", "url": "https://xiyu.news/events/evt_48dd263d7b649072/", "json": "https://xiyu.news/api/events/evt_48dd263d7b649072.json", "type": "product_release", "status": "monitoring", "category": "technology", "title": { "zh": "谷歌发布Gemini 4 Argon模型，四个智能体自主通关《我的世界》", "en": "Google releases Gemini 4 Argon model; four agents autonomously beat Minecraft" }, "current_state": { "zh": "Google DeepMind 让四个 Gemini 4 Argon 编程智能体在《我的世界》中自主规划并协作，经过 27 小时、605 次死亡和约 520 美元的 API 成本后击败末影龙，全程使用自行编写的 JavaScript 而非预置游戏工具。", "en": "Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools." }, "first_seen_at": "2026-10-01T00:03:28.262995+00:00", "last_updated_at": "2026-10-17T08:44:00.376102+00:00", "last_material_change_at": "2026-10-17T08:44:00.376102+00:00", "confidence": 0.75, "updates_count": 5, "sources_count": 7, "entities": [ "argon", "astra", "gemini", "gemini-4", "google", "gpt", "opus", "rsi" ], "identifiers": [ "gpt-6" ], "topics": [ "ai-evaluation", "ai-labs", "ai-pricing", "ai-release", "api-pricing", "benchmarks", "coding", "embodied-ai", "foundation-models", "frontier-model", "frontier-models", "gemini", "gemini-4", "google", "google-deepmind", "google-gemini", "long-horizon-tasks", "model-release", "multi-agent-systems", "robotics" ], "updates": [ { "update_id": "upd_2783cc6655fe9f7f", "event_id": "evt_48dd263d7b649072", "occurred_at": "2026-09-30T23:43:07Z", "published_at": "2026-09-30T23:43:07Z", "first_seen_at": "2026-10-01T00:03:28.262995Z", "time_precision": "published", "update_type": "initial", "material_change": true, "title_zh": "谷歌发布 Gemini 4 Argon，称其为迄今最强大的模型", "title_en": "Google releases Gemini 4 Argon, called its most powerful model yet", "what_changed_zh": "谷歌于 2026 年 9 月 30 日（周三）发布 Gemini 4 Argon，将其宣传为迄今最强大的模型，并定位为面向编程与网络安全工作的主力模型。据 CNBC 报道，该模型首先面向选定的网络安全合作伙伴推出，而非全面开放。\n\n这是谷歌在前沿模型竞争中的最新一次发布；据 CNBC 报道，其初始投放仅面向选定的网络安全合作伙伴，而非所有开发者。谷歌将其定位为开发者、专业人士与企业解决难题的伙伴。\n\n谷歌博客称，Gemini 4 Argon 在编程、知识工作、网络安全防御和创意写作方面具备“前沿级能力”。第三方评测机构 Artificial Analysis 将“Gemini 4 Argon (High)”列为智能水平领先的模型之一，并称其相对同类模型定价合理。", "what_changed_en": "Google released Gemini 4 'Argon', which it markets as its most powerful model to date and pitches as a workhorse for coding and cybersecurity tasks.", "current_state_zh": "谷歌于 2026 年 9 月 30 日（周三）发布 Gemini 4 Argon，将其宣传为迄今最强大的模型，并定位为面向编程与网络安全工作的主力模型。据 CNBC 报道，该模型首先面向选定的网络安全合作伙伴推出，而非全面开放。\n\n这是谷歌在前沿模型竞争中的最新一次发布；据 CNBC 报道，其初始投放仅面向选定的网络安全合作伙伴，而非所有开发者。谷歌将其定位为开发者、专业人士与企业解决难题的伙伴。\n\n谷歌博客称，Gemini 4 Argon 在编程、知识工作、网络安全防御和创意写作方面具备“前沿级能力”。第三方评测机构 Artificial Analysis 将“Gemini 4 Argon (High)”列为智能水平领先的模型之一，并称其相对同类模型定价合理。", "current_state_en": "Google released Gemini 4 'Argon', which it markets as its most powerful model to date and pitches as a workhorse for coding and cybersecurity tasks.", "detailed_summary_zh": "Google released Gemini 4 'Argon', which it markets as its most powerful model to date and pitches as a workhorse for coding and cybersecurity tasks.", "detailed_summary_en": "Google released Gemini 4 'Argon', which it markets as its most powerful model to date and pitches as a workhorse for coding and cybersecurity tasks.", "background_zh": "据 9to5Google 报道，Gemini 4 Argon 采用了新的命名体系，此前谷歌取消了 Gemini 3.5 Pro，并将重心放在 Gemini 3.8 Flash 上。前沿模型通常基于海量数据训练，成本极高，涵盖数据获取与处理以及算力开支。第三方对比报道将 Gemini 4 Argon 与 OpenAI 的 GPT-6 Astra 视为同代竞争对手。", "background_en": "Gemini 4 Argon uses a new naming scheme and follows the cancellation of Gemini 3.5 Pro, with Google having focused on Gemini 3.8 Flash, according to 9to5Google. Frontier models are typically trained on vast datasets at very high cost, covering data acquisition and processing plus compute. Third-party comparison coverage places Gemini 4 Argon alongside OpenAI's GPT-6 Astra as a same-generation rival.", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "该发布与链上资产、托管或代币供给没有直接关联；对加密市场的可能传导渠道是情绪面，即去中心化算力、AI 代理等 AI 叙事类资产历来会对头部实验室的前沿模型发布作出反应。由于模型初期仅面向选定合作伙伴开放，目前尚不能确立开发者需求层面的即时变化。", "market_impact_en": "The release has no direct on-chain, custody or token-supply linkage; any crypto read-through runs through sentiment in AI-narrative assets such as decentralized-compute and AI-agent tokens, which have historically reacted to frontier-model announcements from major labs. Because availability is initially gated to select partners, no immediate developer-demand shift is established.", "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://www.cnbc.com/2026/09/30/google-gemini-4-argon-ai.html", "title": "Google rolls out Gemini 4 Argon, its most advanced model" }, { "url": "https://9to5google.com/2026/09/30/gemini-4-argon-announcement/", "title": "Google announces Gemini 4 Argon as its new frontier model" } ], "confidence": 0.75, "story_ids": [ "rss:techcrunch.com_category_artificial-intelligence_feed_:5e1da0f6c91bb863", "rss:www.qbitai.com_feed:394c1f5a0233310a", "rss:www.theverge.com_rss_ai-artificial-intelligence_index.xml:f0f70ba2683a866f" ], "sources": [ { "url": "https://techcrunch.com/2026/09/30/google-releases-gemini-4-argon-called-its-most-powerful-model-yet/", "label": "TechCrunch AI", "source_type": "rss", "official": false }, { "url": "https://www.qbitai.com/2026/10/499663.html", "label": "QbitAI 量子位", "source_type": "rss", "official": false }, { "url": "https://www.theverge.com/ai-artificial-intelligence/1009862/google-releases-gemini-4-argon", "label": "The Verge AI", "source_type": "rss", "official": false } ] }, { "update_id": "upd_c20fbe06b0208bb0", "event_id": "evt_48dd263d7b649072", "occurred_at": "2026-10-01T00:22:02Z", "published_at": "2026-10-01T00:22:02Z", "first_seen_at": "2026-10-01T02:20:47.462086Z", "time_precision": "published", "update_type": "confirmation", "material_change": true, "title_zh": "Google重回第一梯队：Gemini 4 Argon追平GPT-6 Astra", "title_en": "Google重回第一梯队：Gemini 4 Argon追平GPT-6 Astra", "what_changed_zh": "Artificial Analysis 的独立评估发现，谷歌的 Gemini 4 Argon 在智能指数上以 53 分与 GPT-6 Astra 持平，其幻觉率远低（15%），API 入门价格也更便宜，但全价成本和答案准确性落后于 Astra。", "what_changed_en": "Artificial Analysis' independent evaluation finds Google's Gemini 4 Argon ties GPT-6 Astra on the Intelligence Index at 53 points with a far lower hallucination rate (15%) and cheaper introductory API costs, though full-price costs and answer accuracy lag Astra.", "current_state_zh": "Artificial Analysis 的独立评估发现，谷歌的 Gemini 4 Argon 在智能指数上以 53 分与 GPT-6 Astra 持平，其幻觉率远低（15%），API 入门价格也更便宜，但全价成本和答案准确性落后于 Astra。", "current_state_en": "Artificial Analysis' independent evaluation finds Google's Gemini 4 Argon ties GPT-6 Astra on the Intelligence Index at 53 points with a far lower hallucination rate (15%) and cheaper introductory API costs, though full-price costs and answer accuracy lag Astra.", "detailed_summary_zh": "Artificial Analysis' independent evaluation finds Google's Gemini 4 Argon ties GPT-6 Astra on the Intelligence Index at 53 points with a far lower hallucination rate (15%) and cheaper introductory API costs, though full-price costs and answer accuracy lag Astra.", "detailed_summary_en": "Artificial Analysis' independent evaluation finds Google's Gemini 4 Argon ties GPT-6 Astra on the Intelligence Index at 53 points with a far lower hallucination rate (15%) and cheaper introductory API costs, though full-price costs and answer accuracy lag Astra.", "background_zh": "", "background_en": "", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "", "market_impact_en": "", "importance_score": 8.0, "references": [], "confidence": 0.75, "story_ids": [ "telegram:theblockbeats:198518" ], "sources": [ { "url": "https://m.theblockbeats.info/flash/369836?from=telegram", "label": "theblockbeats", "source_type": "telegram", "official": false } ] }, { "update_id": "upd_83f8f694e53b30b9", "event_id": "evt_48dd263d7b649072", "occurred_at": "2026-10-13T16:03:16Z", "published_at": "2026-10-13T16:03:16Z", "first_seen_at": "2026-10-13T18:21:39.928148Z", "time_precision": "published", "update_type": "confirmation", "material_change": true, "title_zh": "Gemini 4 Argon", "title_en": "Gemini 4 Argon", "what_changed_zh": "Google DeepMind 发布了 Gemini 4“Argon”，这是一款新的前沿模型，可通过 API 使用，入门价格为每百万输入 token 2 美元、每百万输出 token 10 美元（缓存输入享 95% 折扣），因其 SVG 生成和自然语言写作质量而受到关注。", "what_changed_en": "Google DeepMind released Gemini 4 'Argon', a new frontier model available via API at introductory pricing of $2 per million input tokens and $10 per million output tokens (95% off cached input), drawing attention for its SVG generation and natural-language writing quality.", "current_state_zh": "Google DeepMind 发布了 Gemini 4“Argon”，这是一款新的前沿模型，可通过 API 使用，入门价格为每百万输入 token 2 美元、每百万输出 token 10 美元（缓存输入享 95% 折扣），因其 SVG 生成和自然语言写作质量而受到关注。", "current_state_en": "Google DeepMind released Gemini 4 'Argon', a new frontier model available via API at introductory pricing of $2 per million input tokens and $10 per million output tokens (95% off cached input), drawing attention for its SVG generation and natural-language writing quality.", "detailed_summary_zh": "Google DeepMind released Gemini 4 'Argon', a new frontier model available via API at introductory pricing of $2 per million input tokens and $10 per million output tokens (95% off cached input), drawing attention for its SVG generation and natural-language writing quality.", "detailed_summary_en": "Google DeepMind released Gemini 4 'Argon', a new frontier model available via API at introductory pricing of $2 per million input tokens and $10 per million output tokens (95% off cached input), drawing attention for its SVG generation and natural-language writing quality.", "background_zh": "", "background_en": "", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "", "market_impact_en": "", "importance_score": 8.5, "references": [ { "url": "https://news.ycombinator.com/item?id=50073879", "title": "Community discussion" } ], "confidence": 0.75, "story_ids": [ "hackernews:story:50073879" ], "sources": [ { "url": "https://deepmind.google/models/gemini/", "label": "737max", "source_type": "hackernews", "official": false } ] }, { "update_id": "upd_b17e1eef863291ca", "event_id": "evt_48dd263d7b649072", "occurred_at": "2026-10-14T00:37:02Z", "published_at": "2026-10-14T00:37:02Z", "first_seen_at": "2026-10-14T02:08:04.407312Z", "time_precision": "published", "update_type": "escalation", "material_change": true, "title_zh": "谷歌Gemini 4 Argon正式开放：看段视频就能教机器人投沙包，连中7次", "title_en": "谷歌Gemini 4 Argon正式开放：看段视频就能教机器人投沙包，连中7次", "what_changed_zh": "Gemini 4 Argon 从仅面向部分网络安全合作伙伴，扩大至向全球所有付费用户正式开放；可在 Gemini App（Google AI Pro/Ultra）、Antigravity（仅 Ultra）和付费 API 使用；API 首发价每百万输入 2 美元、输出 10 美元，优惠持续至 11 月 30 日后翻倍。谷歌还展示了其视频驱动机器人技能学习和仿真生成能力，ERQA 得分 87%。", "what_changed_en": "Gemini 4 Argon expanded from limited availability for selected cybersecurity partners to a general opening for all global paid users; it is available in the Gemini App (Google AI Pro/Ultra), Antigravity (Ultra only), and paid API. Introductory API pricing of $2/$10 per million tokens lasts until Nov 30, after which it doubles. Google also showcased video-driven robot skill learning, simulation generation, and an 87% ERQA score.", "current_state_zh": "Google DeepMind 的 Gemini 4 Argon 已向全球付费用户开放，此前仅面向部分网络安全合作伙伴。用户可通过 Gemini App（Google AI Pro/Ultra）、Antigravity（仅 Ultra）及付费 API 使用；API 首发价为每百万输入 token 2 美元、输出 10 美元，优惠至 11 月 30 日后翻倍。谷歌同时展示其视频驱动机器人技能学习、自动仿真生成等能力，ERQA 物理推理评测得分 87%。", "current_state_en": "Google DeepMind's Gemini 4 Argon is now open to global paid users after initially being limited to selected cybersecurity partners. It is available via the Gemini App (Google AI Pro/Ultra), Antigravity (Ultra only), and paid API; introductory API pricing is $2 per million input tokens and $10 per million output tokens until Nov 30, after which it doubles. Google also showcased video-driven robot skill learning, automated simulation generation, and an 87% score on the ERQA physical-reasoning benchmark.", "detailed_summary_zh": "Google has opened Gemini 4 Argon to all global paid users, showcasing video-driven robot skill learning (Apptronik's Apollo landing 7 straight sandbag throws), automated simulation generation, 87% on the ERQA physical-reasoning benchmark, and launch API pricing of $2/$10 per million tokens that doubles after November 30.", "detailed_summary_en": "Google has opened Gemini 4 Argon to all global paid users, showcasing video-driven robot skill learning (Apptronik's Apollo landing 7 straight sandbag throws), automated simulation generation, 87% on the ERQA physical-reasoning benchmark, and launch API pricing of $2/$10 per million tokens that doubles after November 30.", "background_zh": "", "background_en": "", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "", "market_impact_en": "", "importance_score": 8.5, "references": [], "confidence": 0.92, "story_ids": [ "telegram:theblockbeats:200329" ], "sources": [ { "url": "https://m.theblockbeats.info/flash/371663?from=telegram", "label": "theblockbeats", "source_type": "telegram", "official": false } ] }, { "update_id": "upd_ddc1081ede1b81a4", "event_id": "evt_48dd263d7b649072", "occurred_at": "2026-10-17T05:33:02Z", "published_at": "2026-10-17T05:33:02Z", "first_seen_at": "2026-10-17T08:44:00.376102Z", "time_precision": "published", "update_type": "confirmation", "material_change": true, "title_zh": "谷歌4个Gemini 4 Argon通关《我的世界》：27小时死605次，花费约520美元", "title_en": "谷歌4个Gemini 4 Argon通关《我的世界》：27小时死605次，花费约520美元", "what_changed_zh": "Google DeepMind 让四个 Gemini 4 Argon 编程智能体在《我的世界》中自主规划并协作，经过 27 小时、605 次死亡和约 520 美元的 API 成本后击败末影龙，全程使用自行编写的 JavaScript 而非预置游戏工具。", "what_changed_en": "Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools.", "current_state_zh": "Google DeepMind 让四个 Gemini 4 Argon 编程智能体在《我的世界》中自主规划并协作，经过 27 小时、605 次死亡和约 520 美元的 API 成本后击败末影龙，全程使用自行编写的 JavaScript 而非预置游戏工具。", "current_state_en": "Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools.", "detailed_summary_zh": "Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools.", "detailed_summary_en": "Google DeepMind let four Gemini 4 Argon coding agents autonomously plan and collaborate in Minecraft, beating the Ender Dragon after 27 hours, 605 deaths and roughly $520 in API costs, using self-written JavaScript rather than pre-built game tools.", "background_zh": "末影龙是《我的世界》的最终 Boss，玩家需先找到要塞、激活传送门才能进入末地；本次实验中一名 Agent 通过两次投掷末影之眼，计算出距目标仅 43 米的位置。Mineflayer 是一个用于构建《我的世界》机器人的开源 JavaScript 框架，Agent 借助它读取坐标、背包与周围环境，寻路则由现成插件负责。根据谷歌开发者文档，Gemini 4 Argon 是面向自主多步骤 Agent 的旗舰推理模型，具备 100 万 Token 上下文窗口；Vals AI 列出的价格为每百万输入 Token 4 美元、每百万输出 Token 20 美元。", "background_en": "The Ender Dragon is Minecraft's final boss, fought in the End dimension after locating a stronghold and activating a portal; in this run one agent threw Eyes of Ender twice to calculate a target about 43 meters away. Mineflayer is an open-source JavaScript framework for building Minecraft bots, which the agents used to read coordinates, inventory and surroundings, while a dedicated plugin handled pathfinding. Gemini 4 Argon is described in Google's developer documentation as a flagship reasoning model built for autonomous multi-step agents with a 1M-token context window, and Vals AI lists it at $4.00 per million input tokens and $20.00 per million output tokens.", "community_discussion_zh": "", "community_discussion_en": "", "market_impact_zh": "本次实验披露的成本与 Token 消耗数据，为长周期 Agent 推理的经济性提供了一个具体参照，这类信息会通过 AI 算力与自主 Agent 叙事影响加密市场的情绪。事件本身不直接涉及具体协议或代币，其传导路径是围绕 Agent 概念及去中心化算力／推理板块的市场情绪，而非文中描述的任何链上资金流动。", "market_impact_en": "The run's disclosed cost and token-consumption data provide a concrete reference point for the economics of long-horizon agentic inference, a channel that feeds the prevailing AI-compute and autonomous-agent narratives in crypto markets. There is no direct exposure to a specific protocol or token; the effect runs through sentiment around agent-themed and decentralized compute/inference segments rather than through any on-chain flow described here.", "importance_score": 7.5, "references": [ { "url": "https://ai.google.dev/gemini-api/docs/models/gemini-4-argon", "title": "Gemini 4 Argon | Gemini API | Google AI for Developers" }, { "url": "https://www.vals.ai/models/google_gemini-4-argon", "title": "Gemini 4 Argon Benchmarks, Cost and Capabilities | Vals AI" }, { "url": "https://minecraft.fandom.com/wiki/Tutorials/Defeating_the_ender_dragon", "title": "Tutorials/Defeating the ender dragon – Minecraft Wiki" } ], "confidence": 0.75, "story_ids": [ "telegram:theblockbeats:201004" ], "sources": [ { "url": "https://m.theblockbeats.info/flash/372343?from=telegram", "label": "theblockbeats", "source_type": "telegram", "official": false } ] } ] }