全部未入选
即梦官网功能监测 · 10-29 23:00 UTC
该线索仅显示即梦官网页面内容可能发生变化，尚未核实是否为新产品发布或新功能上线。
AI 判断：仅为未核实的官网页面变化线索，内容多为既有AI视频与绘画功能介绍，缺乏可演示的新功能或明确时间点。
展开原始摘录
这是官网内容变化线索，观察时间不是产品发布日期；尚未核实是否为新功能，不应直接称为今日发布。
即刻造梦 前往即梦 水母在海底遨游 即梦成片 此视频内容由 即梦AI 生成 支持文/图 生视频 灵感即刻成片 AI视频生成，输入简单的文案或图片，即可快速生成优质视频片段 流畅运镜，生动自然 视频动效效果连贯性强、流畅自然，可轻松操控运镜，调节速度变化，即梦AI为视频智能创作增添无限可能 立即创作 01 从首帧到尾帧， 精准掌控 创新的首帧图片和尾帧图片输入方式，增强视频生成的可控性，轻松打造高品质素材，提升你的视频内容创作效率 立即创作 02 中文创作，得心应手 即梦AI支持根据中文提示词进行创作，拥有更好的语义理解能力，准确把握你的需求，将抽象的思路转化为视觉作品 立即创作 03 支持文/图生图片 AI绘画 梦境成真 即梦AI绘画，仅需简单提示词，即可生成精彩的图片。 你还可以对现有图片进行创意改造，自定义保留人物或主体的形象特征，实现背景替换、风格联想、画风保持、姿势保持等 操作，满足各种场景的创作需求 3D游戏人物风格，戴夸张耳饰的民族风少女 即梦成片 插画风格，太空飞行器在粉色星球低空飞行 即梦成片 未来风，女性机械人形AI 即梦成片 超写实插画，夜晚月光下的花园 即梦成片 文艺复兴风格，身着贵族丝绸服饰猫咪 即梦成片 电影质感，盔甲勇士在火光下奔赴战场 即梦成片 即梦成片 智能画布 多图AI融合 即梦AI一站式智能画布，集成AI拼图生成能力，并提供局部重绘、一键扩图、图像消除和抠图等多功能强大操作。你可以在同一画布上实现多元素的无缝拼接，确保AI绘画的创作风格统一和谐。 立即创作 多图层编辑 精细化控制 连续包年 限时4.5折 连续包月 4.8折 连续包季 限时4.3折 单月购买 基础会员 ¥ 393 每年 首年6折¥393 · 下年续费金额¥659 · 可随时取消 725 积分每月 换算¥10=221积分 立即体验 ✓ Seedance 2.0 VIP 专属通道 ✓ 标准生成通道 ✓ 生成作品去除品牌水印 ✓ 查看全部权益 标准会员 ¥ 1099 每年 首年5.8折¥1099 · 下年续费金额¥1899 · 可随时取消 2210 积分每月 换算¥10=241积分 立即体验 ✓ Seedance 2.0 VIP 专属通道 ✓ 标准生成通道 ✓ 生成作品去除品牌水印 ✓ 查看全部权益 Seedance 2.5 720P限时低至￥0.43/秒 高级会员 限时5折 ¥ 5199 每年 首年5折¥5199 · 下年续费金额¥10398 · 可随时取消 6.2K 12.3K 18.5K 27.7K 12320 积分每月 换算¥10=284积分 立即体验 专享权益 备战双11 首3月16020积分/月 首3月加赠后16020积分/月！SD 2.5 720P带参考视频消耗6.2折 ￥0.43/秒，CLI调用折上9折 ￥0.38/秒 (10月13日-10月30日) ✓ Seedance 2.0 VIP 专属通道 ✓ 图片4.0 4K 限时免费 ✓ 快速生成通道 ✓ 生成作品去除品牌水印 ✓ 查看全部权益 Seedance 2.5 720P限时低至￥0.19/秒 超级会员 限时4.5折 ¥ 19656 每年 首年4.5折¥19656 · 下年续费金额¥43680 · 可随时取消 54600 积分每月 换算¥10=333积分 立即体验 专享权益 备战双11 首3月96060积分/月 1)多得2个月会员；2)首3月得96060积分/月；3)SD2.5 720P带参考视频消耗5折，CLI调用折上9折 ￥0.18/秒 (10月13日-10月30日) ✓ Seedance 2.0 VIP 专属通道 ✓ 图片4.0 4K 限时免费 ✓ 快速生成通道 ✓ 生成作品去除品牌水印 ✓ API服务购买 ✓ 查看全部权益 连续包月 基础会员 ¥ 33 每月 首月4.8折¥33 · 下月续费金额¥69 · 可随时取消 725 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月725积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 标准会员 ¥ 96 每月 首月4.9折¥96 · 下月续费金额¥199 · 可随时取消 2210 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月2210积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 高级会员 6.2K ¥ 259 每月 首月5.2折¥259 · 下月续费金额¥499 · 可随时取消 6160 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月6160积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 12.3K ¥ 519 每月 首月5.3折¥519 · 下月续费金额¥998 · 可随时取消 12320 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月12320积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 18.5K ¥ 779 每月 首月5.3折¥779 · 下月续费金额¥1498 · 可随时取消 18480 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月18480积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 27.7K ¥ 1168 每月 首月5.3折¥1168 · 下月续费金额¥2246 · 可随时取消 27720 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月27720积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 超级会员 ¥ 2235 每月 首月5.2折¥2235 · 下月续费金额¥4299 · 可随时取消 54600 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 API服务购买 购买API套餐且持续为会员状态时，可享受API服务，使用额度需单独购买，与会员积分不通用。 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月54600积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 连续包季 基础会员 ¥ 188 每季 每月¥62.66 · 自动续订，可随时取消 725 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月725积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 标准会员 ¥ 568 每季 每月¥189.33 · 自动续订，可随时取消 2210 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月2210积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 高级会员 6.2K Seedance 2.5 720P限时低至￥1.21/秒 ¥ 1399 每季 每月¥466.33 · 自动续订，可随时取消 6160 积分每月 专享权益 Seedance 2.5 720P带参考视频消耗6.2折，低至￥1.21/秒 (10月13日-10月30日) Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月6160积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 12.3K Seedance 2.5 720P限时低至￥0.45/秒 限时4.5折 ¥ 1259 每季 首季4.5折¥1259 · 下季续费金额¥2798 · 可随时取消 12320 积分每月 专享权益 首月加赠2460积分，共得14780积分！再享Seedance 2.5 720P带参考视频消耗6.2折，低至￥0.45/秒 (10月13日-10月30日) 首月14780积分 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外赠送积分每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月12320积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 18.5K Seedance 2.5 720P限时低至￥0.4/秒 限时4.5折 ¥ 1890 每季 首季4.6折¥1890 · 下季续费金额¥4199 · 可随时取消 18480 积分每月 专享权益 首月加赠6470积分，共得24950积分！再享Seedance 2.5 720P带参考视频消耗6.2折，低至￥0.4/秒 (10月13日-10月30日) 首月24950积分 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外赠送积分每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月18480积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 27.7K Seedance 2.5 720P限时低至￥0.39/秒 限时4.5折 ¥ 2834 每季 首季4.5折¥2834 · 下季续费金额¥6298 · 可随时取消 27720 积分每月 专享权益 首月加赠11080积分，共得38800积分！再享Seedance 2.5 720P带参考视频消耗6.2折，低至￥0.39/秒 (10月13日-10月30日) 首月38800积分 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外赠送积分每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月27720积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 超级会员 Seedance 2.5 720P限时低至￥0.27/秒 限时4.3折 ¥ 4999 每季 首季4.3折¥4999 · 下季续费金额¥11699 · 可随时取消 54600 积分每月 专享权益 首月加赠26000积分，共得80600积分！再享Seedance 2.5 720P带参考视频消耗5折，低至￥0.27/秒 (10月13日-10月30日) 首月80600积分 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外赠送积分每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 API服务购买 购买API套餐且持续为会员状态时，可享受API服务，使用额度需单独购买，与会员积分不通用。 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月54600积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 单月购买 基础会员 ¥ 79 每月 一次性支付¥79，有效期1个月，不自动续费 725 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月725积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 标准会员 ¥ 239 每月 一次性支付¥239，有效期1个月，不自动续费 2210 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月2210积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 高级会员 ¥ 649 每月 一次性支付¥649，有效期1个月，不自动续费 6160 积分每月 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月6160积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 连续包年 基础会员 ¥ 393 每年 首年6折¥393 · 下年续费金额¥659 · 可随时取消 725 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月725积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 标准会员 ¥ 1099 每年 首年5.8折¥1099 · 下年续费金额¥1899 · 可随时取消 2210 积分每月 Seedance 2.0 VIP 专属通道 标准生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月2210积分 seedance2.0 VIP模型：✓ 视频生成专享加速：标准生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 智能超清：4K 去除品牌水印：✓ 生成加速：✓ 高级会员 6.2K Seedance 2.5 720P限时低至￥0.47/秒 限时5折 ¥ 2599 每年 首年5折¥2599 · 下年续费金额¥5199 · 可随时取消 6160 积分每月 专享权益 首3月加赠后7410积分/月！SD 2.5 720P带参考视频消耗6.2折 ￥0.47/秒，CLI调用折上9折 ￥0.41/秒 (10月13日-10月30日) 首3月7410积分/月 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外积分连续加赠3个月，每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月6160积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 12.3K Seedance 2.5 720P限时低至￥0.43/秒 限时5折 ¥ 5199 每年 首年5折¥5199 · 下年续费金额¥10398 · 可随时取消 12320 积分每月 专享权益 首3月加赠后16020积分/月！SD 2.5 720P带参考视频消耗6.2折 ￥0.43/秒，CLI调用折上9折 ￥0.38/秒 (10月13日-10月30日) 首3月16020积分/月 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外积分连续加赠3个月，每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月12320积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 18.5K Seedance 2.5 720P限时低至￥0.36/秒 限时4.5折 ¥ 7019 每年 首年4.5折¥7019 · 下年续费金额¥15598 · 可随时取消 18480 积分每月 专享权益 首3月加赠后25880积分/月！SD 2.5 720P带参考视频消耗6.2折 ￥0.36/秒，CLI调用折上9折 ￥0.32/秒 (10月13日-10月30日) 首3月25880积分/月 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外积分连续加赠3个月，每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月18480积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 高级会员 27.7K Seedance 2.5 720P限时低至￥0.34/秒 限时4.5折 ¥ 10529 每年 首年4.5折¥10529 · 下年续费金额¥23398 · 可随时取消 27720 积分每月 专享权益 首3月加赠后41620积分/月！SD 2.5 720P带参考视频消耗6.2折 ￥0.34/秒，CLI调用折上9折 ￥0.30/秒 (10月13日-10月30日) 首3月41620积分/月 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外积分连续加赠3个月，每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月27720积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 超级会员 Seedance 2.5 720P限时低至￥0.19/秒 限时4.5折 ¥ 19656 每年 首年4.5折¥19656 · 下年续费金额¥43680 · 可随时取消 54600 积分每月 专享权益 1)多得2个月会员；2)首3月得96060积分/月；3)SD2.5 720P带参考视频消耗5折，CLI调用折上9折 ￥0.18/秒 (10月13日-10月30日) 首3月96060积分/月 1. 加赠福利为大型促销活动，暂不支持退款。法律法规另有规定除外。 2. 额外积分连续加赠3个月，每月跟随订阅积分发放，非一次性发放，有效期为29天。3. 特别注意：每个账户仅能享受1次加赠福利，多次购买不叠加，退款后重新购买不重新发放。一旦会员失效，对应赠送的福利即时终止。4. 单秒价格计算方式为：模型1秒消耗积分数*（活动价/订阅周期/获得积分数），所得结果四舍五入保留两位小数 Seedance 2.0 VIP 专属通道 图片4.0 4K 限时免费 持续为高级会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 快速生成通道 生成作品去除品牌水印 API服务购买 购买API套餐且持续为会员状态时，可享受API服务，使用额度需单独购买，与会员积分不通用。 平台免费积分：每日登录赠送积分 充值购买积分：✓ 订阅会员积分：每月54600积分 seedance2.0 VIP模型：✓ 视频生成专享加速：快速生成通道 视频对口型：✓ 视频高清：✓ 视频补帧：✓ 图片 4.0 限时免费：4K 持续为会员状态时，可享受图片 4.0 4k生成免费至2026.11.06 智能超清：4K/8K 去除品牌水印：✓ 生成加速：✓ 创意涌动 灵感绽放 在即梦AI的创意社区，与其他用户共同探索无限的影像灵感。让创意在相互碰撞中激发，一起创造精彩 获取创作灵感 🍰Berry 16 创作同款 盒饭饭喜欢吃丸子 26 创作同款 小树树爱世界！ 56 创作同款 晓梦 35 创作同款 公主 34 创作同款 塔尼雅 10 创作同款 不吃草的 花牛 25 创作同款 水水 19 创作同款 晗槑槑 17 创作同款 🍰Berry 10 创作同款 檬檬 5 创作同
API credits benefit for eligible ChatGPT subscriptions
Simon Willison · 10-29 22:48 UTC
标题称，符合条件的 ChatGPT 订阅用户可获得 API credits 权益。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是面向开发者的 API 额度权益，普通用户相关性有限，且缺少可核实的使用细节。
展开原始摘录
API credits benefit for eligible ChatGPT subscriptions
Earlier this month Anthropic
announced a generous new benefit
for their Max and Team subscriptions: $100, $200, and $500 per month subscribers would get the equivalent quantity of (expiring) credits to use with the Anthropic API.
OpenAI have now launched pretty much exactly the same deal: if you have a $100, $200, or $500 subscription plan with them you now get that same quantity of credits to apply to one of your API organizations.
I maintain subscriptions with both companies, so this is great news for me! I may even ship some public-facing demos now that the credits are already covered by my existing subscriptions.
Tags:
ai
,
openai
,
generative-ai
,
llms
,
anthropic
,
llm-pricing
Tiberium · 10-29 21:38 UTC
Binance-hacker-1029 – $5.6B stolen from Binance
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[scrlk]: Their insurance fund (SAFU) is completely wiped out, it's only $1 billion in BTC: https://www.binance.com/en/academy/glossary/secure-asset-fun... The name of the fund was originally inspired by this meme: https://www.youtube.com/watch?v=DelF6zEHXpE FWIW, at the end of every crypto bear market, a CEX usually blows up (eg FTX)...
[Tiberium]: About $2.4 billion out of this is already frozen by Circle and Tether. Some of the big ones: - 971.15M USDT (Ethereum): https://etherscan.io/tx/0x7baba482077843514d9c72bac0110ba091... - 823M USDC (Ethereum): https://etherscan.io/tx/0xc954c86f118e8e410e6a3eb9e3b38ff4a9... - 312.73M USDT (Tron): https://tronscan.org/#/transaction/71b0ec0fe873fe820c37cc3ec... - 175M USDT (Ethereum): https://etherscan.io/tx...
[glimshe]: At least they are safe from governments and their "fiat money"!
[ChoosesBarbecue]: Molly White (she runs web3isgoinggreat) seems to think this is real [0] [0]: https://bsky.app/profile/molly.wiki/post/3mz27m5slcc2o
[darkwi11ow]: The hack is publicized for at least a hour, but from reporting I saw they still continue to drain funds as for now. I wonder why large CEX do not have any safety measures for this kind of hacks, they could withdraw everything to newly generated wallets and minimize damage that way. It is certainly not the first time CEX wallets get hacked.
Reddit Shifts to Logged-In Experience as AI Search Kills Open-Web Traffic
Future Tools · 10-29 21:00 UTC
Reddit 宣布将主要转向登录后体验，因为 AI 搜索导致站外搜索推荐流量大幅下降，并可能影响下一季度日活用户。
AI 判断：这是一则有具体数据支撑的行业趋势，但对普通人的直接可用性有限，更适合解读 AI 搜索对开放网络的影响。
展开原始摘录
Company News · 2h ago Reddit Shifts to Logged-In Experience as AI Search Kills Open-Web Traffic Reddit is shifting to a primarily logged-in experience as AI-powered search decimates open-web traffic referrals. CEO Steve Huffman said search referrals dropped over 1 million per week in October, and daily active users could fall by a low-double-digit percentage next quarter from 130.3 million. The move aims to protect Reddit's data from AI scraping while enabling licensing deals and better ad targeting. Despite user declines, Q3 revenue rose 53% year-over-year to $895 million.
Figma AI / Agent / Make · 10-29 21:00 UTC
Figma 发文介绍其训练专用画布编辑模型的方法与思路。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是模型训练方法的工程分享，面向开发者，普通用户缺少可直接使用的场景与演示。
展开原始摘录
Figma 介绍其为 agent 自研 Figma 编辑模型的训练方法，先用超过 25 万条合成样本做 SFT，让早期 Llama 3.3 70B 在编辑任务上接近当时顶级模型，后转向 RL 并结合确定性校验器与视觉评委做评测。
TechCrunch：AI（RSS） · 10-29 20:08 UTC · 同链接合并评分
Reddit 宣布将主要转向登录后体验，因为 AI 搜索导致站外搜索推荐流量大幅下降，并可能影响下一季度日活用户。
AI 判断：这是一则有具体数据支撑的行业趋势，但对普通人的直接可用性有限，更适合解读 AI 搜索对开放网络的影响。
展开原始摘录
Reddit 在 Q3 财报中宣布将转向以登录为主的体验，以保护数据免受 AI 抓取并提升广告变现。10 月搜索流量平均每周减少超 100 万次引荐，Q3 全球日活为 1.303 亿，其中美国日活降至 5070 万，同比降 2%。Q3 收入同比增长 53% 至 8.95 亿美元，净利润 2.83 亿美元；该转变将先在美国以外市场测试，暂无全球上线时间表。
Meta Launches Style Sketch AI Fashion App with Instagram Closet Integration
Future Tools · 10-29 20:00 UTC
Meta 在印尼、新加坡和泰国上线 iOS 与 Android 版 Style Sketch，能把 Instagram 照片变成虚拟衣橱，并生成搭配方案、虚拟试穿和按目的地天气安排的旅行穿搭计划。
AI 判断：这是普通人可直接上手的消费级 AI 功能，画面演示感强，且有明确的国家与平台限制信息，但实际可用范围仅限三国。
展开原始摘录
Tools & Products · 3h ago Meta Launches Style Sketch AI Fashion App with Instagram Closet Integration Meta has quietly launched Style Sketch, an AI fashion app available in Indonesia, Singapore, and Thailand on iOS and Android. The app connects to Instagram to build a virtual closet from posted photos, or lets users add items manually via photo, video, or camera roll uploads. Style Sketch suggests outfit combinations, lets users virtually try them on a realistic model matched to their body shape and skin tone, and can generate day-by-day trip outfit plans based on destination, weather, and activities.
openai/codex released rust-v0.166.0
github-actions[bot] · 10-29 19:47 UTC
OpenAI Codex 的 Rust 版命令行工具发布 v0.166.0，更新键盘选择、Markdown 渲染、模型目录和队列恢复等功能。
AI 判断：这是面向开发者的命令行工具常规版本更新，普通用户难以直接使用，且模型目录新增不构成独立模型发布。
展开原始摘录
## New Features
- Added keyboard-driven transcript selection with word, line, and page motions, integrated search, and configurable shortcuts. (#57221, #57222, #57223, #57224)
- Rendered Markdown consistently in async questions, choices, answers, and transcript replies. (#57259, #57292)
- Added GPT-6.1 Terra to the bundled, Vertex AI, and Amazon Bedrock model catalogs. (#57293)
- Added `codex queue --resume` to resume unloaded threads after queueing a message. (#57232)
- Added clearer safety-block explanations and selectable safer follow-up prompts in the TUI. (#57363, #57365)
- Added `codex doctor --network-only` for focused configuration, connectivity, and WebSocket diagnostics. (#57606)
## Bug Fixes
- Preserved acknowledged approval and permission settings when resuming or forking threads. (#57227)
- Preserved local image attachments when restoring killed composer text. (#57502)
- Fixed standalone release cleanup on macOS when installations are in use or paths contain non-ASCII characters. (#57503)
- Restored spawned descendant threads when unarchiving their parent. (#57635)
- Fixed app-server thread startup and shutdown races and released completed runtimes promptly. (#57787, #57788)
- Improved TUI reconnection recovery, including offline `/quit` and `/exit` handling. (#57820)
- Preserved reasoning-token accounting across resume and fork to avoid premature compaction. (#57839)
## Chores
- Switched Bazel to source-built V8 with target-matched Rust bindings and native Windows MSVC archive support. (#57164, #57269, #57644, #57832)
- Shipped the standalone executor in canonical CLI and app-server release packages. (#57837)
- Standardized development and release builds on Node.js 24.22.1 and Rust 1.95.0. (#57168, #57373)
## Changelog
Full Changelog: https://github.com/openai/codex/compare/rust-v0.165.0...rust-v0.166.0
- #56468 Preserve `SystemRoot` for Windows filesystem sandbox helpers @jif-oai
- #56477 Move Guardian approval reason caps into context composition @ioanbudea-OAI
- #56483 Allow installed plugin metadata to skip display enrichment @bkotsopoulos
- #56493 Move Guardian tool description preparation into context composition @ioanbudea-OAI
- #56494 Add channel upserts to message boards @felixxia-oai
- #56495 Add paginated channel subscription queries to the local message board @felixxia-oai
- #56505 Defer synchronous Guardian tool-call truncation to preparation @ioanbudea-OAI
- #56514 Preserve tracing context throughout tool dispatch @bkotsopoulos
- #56515 Add trace instrumentation for turn startup and response streaming @bkotsopoulos
- #56516 Respect server approval policy defaults during TUI startup @etraut-openai
- #56524 Add turn, tool, and skill attribution to traces @bkotsopoulos
- #56525 Remove the 1 MiB response limit for explicit model catalogs @eternal-openai
- #56530 Record skill and plugin catalog sizes on publication @TAFOYA-OAI
- #56533 Defer sync Guardian tool-output truncation to transcript preparation @ioanbudea-OAI
- #56535 Extract the head-tail buffer into a utility crate @seanh-oai
- #56543 Track resume picker pagination and selection metadata per backend @bc-openai
- #56552 Initialize SQLite databases with DELETE journaling on NFS @dkovalenko-oai
- #56553 Include environment hooks in thread hook listings @andrewgu-oai
- #56554 Enable masked AWS credentials for local proxy signing @viyatb-oai
- #56555 Apply sync Guardian message limits during transcript preparation @ioanbudea-OAI
- #56556 Use conversations for all Guardian v2 async classification @felixxia-oai
- #56557 Exercise retained sampling in Guardian transport tests @felixxia-oai
- #56559 Remove Guardian snapshot-specific context helpers @felixxia-oai
- #56560 Allow configuring GCP project and location in provider settings @alexsong-oai
- #57164 Move the V8 Bazel build graph into Codex @seanh-oai
- #57165 Enforce V8 artifact profiles at Bazel target boundaries @seanh-oai
- #57166 Surface automatic marketplace update failures as startup warnings @dkovalenko-oai
- #57167 Track CUA tool call starts and deduplicate lifecycle events @trpcic-openai
- #57168 Centralize Node.js version selection on 24.22.1 @anp-oai
- #57169 Clarify config loader APIs and share interactive login setup @smurching-oai
- #57170 Update prebuilt argument-comment-lint to `rust-v0.164.0` @tamird
- #57171 Integrate runtime authentication into model requests @cooper-oai
- #57172 Expose lifecycle timestamps on app-server thread items @charliemarsh-oai
- #57174 Add scoped request-body secret substitution to MITM hooks @viyatb-oai
- #57175 Integrate config-server policies with startup and authentication @smurching-oai
- #57176 Defer Guardian Node REPL output truncation until prompt preparation @ioanbudea-OAI
- #57178 Route FedRAMP API keys to government endpoints by default @lsingh-oai
- #57179 Consolidate async header authentication in `apply_auth_headers` @cooper-oai
- #57180 Add constraint-aware standalone network proxy configuration @viyatb-oai
- #57181 Add optional names to untrusted additional context @akira-oai
- #57185 Centralize unified exec output rendering @anp-oai
- #57186 Fail closed on unreadable glob search root metadata errors @viyatb-oai
- #57187 Allow advisory rollout budgets with optional limit enforcement @nvogt-oai
- #57188 Add a credential file override for GCP provider authentication @alexsong-oai
- #57191 Read unified exec output directly from stdout and stderr @anp-oai
- #57192 Gate capability discovery V2 on executor version @TAFOYA-OAI
- #57193 Preserve failed `write_stdin` calls in rollout trace replay @cassirer-openai
- #57194 Classify rollout tool calls by their runtime handler @cassirer-openai
- #57196 Isolate MCP inputs and timeouts for callers sharing a connection @charliemarsh-oai
- #57197 Clarify skill toggle summaries in the TUI @fcoury-oai
- #57198 Report the local command working directory in PreToolUse hooks @eternal-openai
- #57199 Truncate rollout JSON summaries at UTF-8 character boundaries @cassirer-openai
- #57201 Capture direct-tool critical paths behind a feature flag @taohong339
- #57202 Preserve the tool dispatch span across runtime futures @trpcic-openai
- #57203 Preserve command cwd in permission and post-tool hooks @eternal-openai
- #57204 Register the `untrusted_as_tool_output` feature flag @akira-oai
- #57206 Ignore the crypto-bigint timing advisory in dependency checks @etraut-openai
- #57207 Attribute approval waits to tool critical-path timing @taohong339
- #57208 Propagate installer download failures in standalone Unix updates @etraut-openai
- #57209 Report turn critical-path timing in analytics @taohong339
- #57210 Add opt-in protocol discovery for local stdio MCP servers @charliemarsh-oai
- #57211 Share eligible local stdio MCP connections across chats @charliemarsh-oai
- #57212 Enable zstd compression for code-mode gRPC responses @cayoub-oai
- #57215 Coordinate resume picker sources independently @bc-openai
- #57218 Support negotiated WebSocket compression in the remote app-server client @thorvald-oai
- #57220 Log app-server WebSocket compression statistics @thorvald-oai
- #57221 Expand keyboard navigation and selection in the transcript @fcoury-oai
- #57222 Integrate Find with transcript selection mode @fcoury-oai
- #57223 Wire keyboard selection into transcript overlays @fcoury-oai
- #57224 Fix transcript selection input routing and preserve composer drafts @fcoury-oai
- #57225 Add managed enforcement of DELETE journaling on NFS @dkovalenko-oai
- #57226 Allow NFS database migration with shared file permissions @dkovalenko-oai
- #57227 Preserve acknowledged permissions when resuming and forking threads @charliemarsh-oai
- #57228 Give remote control pairing a 35-second deadline @etraut-openai
- #57229 Simplify code-mode cell completion delivery @cconger
- #57232 Add optional thread resumption to `codex queue` @etraut-openai
- #57233 Stop dotenv loading on persistent I/O errors @etraut-openai
- #57234 Prevent credential collisions in registry auth test fixtures @jif-oai
- #57235 Tag exec_command telemetry with the configured shell snapshot mode @jif-oai
- #57238 Exercise V2 spawning in subagent context inheritance tests @tamird
- #57239 Make the MCP HTTP probe timeout test deterministic @jif-oai
- #57240 Bound memory baseline files and recover incomplete snapshots @dhruvgupta-oai
- #57241 Honor configured message limits for encrypted guardian transcripts @ioanbudea-OAI
- #57242 Reduce the default thread unload delay to one minute @jif-oai
- #57243 Pause Tokio time in startup draft handoff tests @goodman-oai
- #57244 Wait for process completion in the unified exec reuse test @goodman-oai
- #57245 Stabilize the rejected queue message test @goodman-oai
- #57247 Release the rollout maintenance lock when its guard is dropped @goodman-oai
- #57248 Make login token-refresh tests wait for auth publication @goodman-oai
- #57249 Allow disabling plugin discovery prefetch at app-server startup @zamoshchin-openai
- #57251 Show native dollar thread usage for ENT26 plans @jif-oai
- #57253 Allow slow daemon startup while detecting early process exits @etraut-openai
- #57255 Record reviewed exceptions for zstd security advisories @charliemarsh-oai
- #57256 Keep parent history out of attached thread views @btraut-openai
- #57257 Avoid unnecessary connector directory work for tool suggestions @charliemarsh-oai
- #57258 Reduce remote plugin catalog cache probing overhead @charliemarsh-oai
- #57259 Render Markdown in TUI async question choices @imac-oai
- #57260 Centralize permission request preparation and approval actions @swastik-oai
- #57261 Separate `apply_patch` request preparation from execution @swastik-oai
- #57262 Add incremental skill catalogs behind `skill_catalog_deltas` @adaley-openai
- #57263 Support model-catalog wording for permission path warnings @goodman-oai
- #57265 Show effective processing tiers in TUI status surfaces @etraut-openai
- #57267 Keep cloud conversation settings server-owned in the TUI @bc-openai
- #57268 Record reasons for embedded TUI starts @etraut-openai
- #57269 Add native Windows MSVC V8 source archive builds @seanh-oai
- #57270 Load cloud config for sandbox and remote exec-server commands @abhinav-oai
- #57271 Enable text selection and mouse scrolling in the warning viewer @imac-oai
- #57272 Add code-mode output truncation protocol contracts @seanh-oai
- #57273 Reject unsupported options for `codex queue` @etraut-openai
- #57274 Defer REPL evidence rendering to synchronous context preparation @ioanbudea-OAI
- #57275 Preserve image generation failure metadata for premium usage limits @jimxie-oai
- #57276 Support window prefixes for extension world-state sections @fc-oai
- #57277 Use the crates.io release of `i18n-config` @goodman-oai
- #57278 Refactor agent spawning, startup, and communication helpers @rhan-oai
- #57279 Render untrusted additional context as named tool outputs @akira-oai
- #57280 Preserve full MCP results in rollout traces @cassirer-openai
- #57281 Enable agents navigation for remote App Server sessions @etraut-openai
- #57282 Promote rounded token counts to the next compact unit @etraut-openai
- #57284 Express TUI mouse capture as an explicit policy @imac-oai
- #57285 Let the terminal handle links in expanded questions @imac-oai
- #57286 Support `network_proxy` in managed requirements @viyatb-oai
- #57287 Preserve parent tool declarations in prediction forks @keyz
- #57290 Add a plugin process host with bounded IPC and lifecycle management @joesavona
- #57291 Represent user message presentation with an enum @imac-oai
- #57292 Render async questions and replies with Markdown in the TUI @imac-oai
- #57293 Add GPT-6.1 Terra to model catalogs @BOZHANG-OAI
- #57294 Extract plugin manager errors and box realtime handoff routing @TAFOYA-OAI
- #57295 Defer synchronous Guardian evidence limits to preparation @ioanbudea-OAI
- #57304 Surface refresh-token rejection errors during workspace routing @smurching-oai
- #57307 Bump `openai/fence` to `v0.12.2` in the blob size policy @joesavona
- #57311 Preserve tool output versions and history order in rollout traces @cassirer-openai
- #57312 Centralize Guardian root context preparation @ioanbudea-OAI
- #57313 Default gateway OAuth delivery to Authorization bearer tokens @smurching-oai
- #57314 Format all Rust sources and clean up repository checks @joesavona
- #57315 Keep failed subagents from marking tasks as needing attention @bc-openai
- #57317 Add opt-in concise hints for explicitly selected plugins @adaley-openai
- #57325 Add model catalog fields for granular approval templates @goodman-oai
- #57342 Separate TUI session resume preparation from activation @bc-openai
- #57352 Preserve child signal failures in the npm wrapper @dkovalenko-oai
- #57353 Avoid object-identity preflight for MXC path grants @iceweasel-oai
- #57357 Report when NFS database migration finds no databases @dkovalenko-oai
- #57358 Serialize session index mutations across processes @dkovalenko-oai
- #57360 Bound waits for the thread writer coordination lock @dkovalenko-oai
- #57363 Explain safety policy blocks with turn-scoped recourse in the TUI @etraut-openai
- #57365 Offer safer rewrites after safety blocks in the TUI @etraut-openai
- #57366 Show the selected ChatGPT workspace in `/status` @etraut-openai
- #57373 Pin the Rust toolchain to 1.95.0 @TAFOYA-OAI
- #57391 Drain pending terminal reports before returning to the shell @etraut-openai
- #57393 Render integral limits and parenthesized matrices in TUI math @etraut-openai
- #57394 Add experimental GCP account setup to the app server @alexsong-oai
- #57420 Make thread item timestamps optional in TypeScript exports @keyz
- #57425 Centralize Guardian root context preparation @ioanbudea-OAI
- #57430 Test goal provenance in existing workers' Guardian reviews @jif-oai
- #57433 Add opt-in synchronous Guardian review for native tool calls @jif-oai
- #57435 Extract unified exec command preparation into a reusable module @swastik-oai
- #57438 Carry prepared commands through unified execution @swastik-oai
- #57439 Preserve Guardian request metadata with custom OpenAI base URLs @jif-oai
- #57441 Add Guardian native multi-agent history feature flag @jif-oai
- #57448 Preserve native multi-agent calls in Guardian review history @jif-oai
- #57451 Route state-changing multi-agent calls through Guardian review @jif-oai
- #57452 Distinguish policy-restricted capacity failures from server overloads @toml-openai
- #57455 Centralize Guardian classification of multi-agent tool calls @jif-oai
- #57458 Move Guardian token usage calculation into the reviewer crate @jif-oai
- #57462 Always put Code Mode tool descriptions before namespace descriptions @swastik-oai
- #57468 Fix guardian handoff mocks for native history tests @jif-oai
- #57474 Test encrypted native call history across Guardian reviews @jif-oai
- #57476 Move Guardian session helpers into the reviewer crate @jif-oai
- #57478 Move Guardian approval request types into the reviewer crate @jif-oai
- #57483 Skip unrelated indexed rollouts during selected thread migration @charliemarsh-oai
- #57488 Move guardian prompt composition into the reviewer crate @jif-oai
- #57493 Move Guardian action JSON formatting into the reviewer crate @jif-oai
- #57494 Centralize synchronous Guardian context preparation @ioanbudea-OAI
- #57500 Add an option to omit descriptions from skill catalogs @adaley-openai
- #57501 Render escaped percent signs in TUI math @etraut-openai
- #57502 Preserve local image attachments when restoring killed text @etraut-openai
- #57503 Fix macOS standalone release pruning @etraut-openai
- #57504 Use the native clipboard for WSL SSH sessions with X11 forwarding @etraut-openai
- #57506 Move Guardian request mappings into the reviewer crate @jif-oai
- #57507 Resolve Guardian command path conventions inside `ReviewAction` @swastik-oai
- #57508 Add a bounded V8 runtime for UI plugin processes @joesavona
- #57509 Extract Guardian planned-action assembly into the reviewer crate @jif-oai
- #57512 Explicitly unlock daemon PID reservation probes on Unix @jif-oai
- #57513 Replace the npm wrapper with the native binary using `execve` @bolinfest
- #57514 Add experimental ChatGPT workspace switching to app-server @dhruvgupta-oai
- #57515 Limit MXC volume expansion probes to symbolic links @iceweasel-oai
- #57516 Add profile and provider status line items @etraut-openai
- #57519 Defer synchronous Guardian heartbeat compression to preparation @ioanbudea-OAI
- #57522 Add isolated reviews to the guardian reviewer pool @swastik-oai
- #57533 Release empty PID probe locks explicitly on Unix @jif-oai
- #57538 Preserve host-assigned user input IDs in Responses messages @zbrunson-openai
- #57542 Add model-gated literature search to standalone web search @rhan-oai
- #57552 Expose and persist image generation IDs @jimxie-oai
- #57553 Allow replacing dynamic tools between model steps @sdcoffey
- #57554 Keep submitted prompts shaded when terminal colors are unavailable @bc-openai
- #57563 Avoid retaining unneeded output items in HTTP response streams @btraut-openai
- #57564 Preserve rich composer content in Vim registers @bc-openai
- #57567 Allow catalog URL overrides for the built-in OpenAI provider @jamy-OAI
- #57568 Configure memory consolidation service tiers independently @mandal-OAI
- #57569 Preserve process output and control replies under backpressure @ztanner-oai
- #57572 Support Apps tool catalogs without shared snapshots @bkotsopoulos
- #57573 Add filesystem-free classification of intercepted patch commands @swastik-oai
- #57575 Refactor HTTP client construction and request execution @bz-openai
- #57581 Allow model catalogs to customize approval instruction headings @goodman-oai
- #57584 Preserve output boundaries through token truncation @seanh-oai
- #57585 Replace unavailable image references in model requests @kchainani-oai
- #57586 Add host-reserved V2 child creation APIs @rhan-oai
- #57589 Add a standalone `codex-exec-server` executable @dkovalenko-oai
- #57590 Ship `codex-exec-server` in CLI and app-server packages @dkovalenko-oai
- #57593 Remove empty local staging routing files @fcoury-oai
- #57597 Preserve WebSocket probe evidence and redact diagnostic details @dob-oai
- #57598 Cache empty remote skill discovery with filesystem watches @adams-oai
- #57599 Support catalog headings for saved command-prefix notices @goodman-oai
- #57606 Add `codex doctor --network-only` for focused network diagnostics @dob-oai
- #57607 Share runtime authentication with model providers and catalogs @cooper-oai
- #57612 Disable terminal status sequences on older VTE backends @fcoury-oai
- #57615 Preserve rollout tracing across sub-agent reloads @cassirer-openai
- #57618 Release sampling prompts after stream setup @btraut-openai
- #57626 Show Bedrock setup for explicitly configured OpenAI providers @alexsong-oai
- #57628 Stop shipping `codex-exec-server` in release artifacts @fcoury-oai
- #57635 Restore spawned descendants when unarchiving a thread @etraut-openai
- #57638 Add OpenTelemetry requirement types to the app-server protocol @arnavpatidar-OAI
- #57644 Generate Rust bindings that match Bazel-built V8 archives @seanh-oai
- #57649 Support managed OpenTelemetry configuration requirements @abhinav-oai
- #57654 Redact OTEL exporter headers in app-server config responses @arnavpatidar-OAI
- #57657 Fix trace exporter activation after account switches @arnavpatidar-OAI
- #57658 Support managed OpenTelemetry metadata and PEM client identities @abhinav-oai
- #57667 Remove detached delivery from `review/start` @yansenzhou-oai
- #57692 Wait for turn profiles before asserting sampling counts in trace tests @ioanbudea-OAI
- #57697 Move Guardian review state into the reviewer extension @jif-oai
- #57698 Move Guardian context policy and prompts into the reviewer extension @jif-oai
- #57702 Move Guardian review evidence into the reviewer extension @jif-oai
- #57710 Preserve parent routing affinity for Guardian reviews @jif-oai
- #57720 Move `ReviewAction` into the Guardian reviewer crate @jif-oai
- #57723 Disable optimization for the on-demand Bazel schema generator @jif-oai
- #57729 Remove `free_guardian` configuration from Guardian tests @ioanbudea-OAI
- #57741 Separate test-only Rust sources from Bazel library and binary inputs @jif-oai
- #57742 Avoid recompiling Bazel integration tests for runtime fixture changes @jif-oai
- #57743 Move Guardian MCP elicitation helpers into the reviewer crate @jif-oai
- #57757 Expose Guardian assessments in response streams @jif-oai
- #57758 Include terminal launch commands in Guardian stdin reviews @jif-oai
- #57770 Respect model catalog origins when selecting Guardian reviewers @ioanbudea-OAI
- #57774 Skip non-regular files when detecting pasted image paths @etraut-openai
- #57782 Restrict code-mode host replacement to session admission @aakash-oai
- #57783 Exclude unrelated network shares from MXC default reads @iceweasel-oai
- #57787 Fix app-server thread listener startup and shutdown races @charliemarsh-oai
- #57788 Release app-server thread runtimes after shutdown @charliemarsh-oai
- #57789 Refresh the model catalog once when Guardian's default reviewer is missing @jif-oai
- #57790 Centralize synchronous Guardian prompt preparation @ioanbudea-OAI
- #57796 Extend GitHub release read retries and add jitter @imac-oai
- #57800 Backfill exec completion items through paginated history @btraut-openai
- #57817 Detach realtime sideband observers without ending the session @jiahao-oai
- #57818 Reduce unnecessary work in configuration processing @bc-openai
- #57819 Load up to eight plugins concurrently @bc-openai
- #57820 Improve TUI reconnect recovery and offline exit handling @etraut-openai
- #57824 Include directly attached threads in diagnostic uploads @btraut-openai
- #57825 Update CI setup to use `meta/install-dotslash` @seanh-oai
- #57832 Build V8 from source by default in Bazel @seanh-oai
- #57834 Add isolated Guardian assessments for prepared actions @swastik-oai
- #57835 Limit concurrent image availability checks to eight @kchainani-oai
- #57836 Declare the image generation result type in code mode @alicec-oai
- #57837 Ship the standalone executor in release packages @dkovalenko-oai
- #57839 Preserve reasoning token accounting across resume and fork @andrewgu-oai
- #57840 Validate thread providers before resume and input @abhinav-oai
- #57841 Add optional process metrics sampling for code-mode hosts @aakash-oai
j4mie · 10-29 19:41 UTC
Patrick Stewart has died
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
https://archive.ph/y4Egm
https://en.wikipedia.org/wiki/Patrick_Stewart
--- Top Comments ---
[halflife]: Watching him in Star Trek TNG made me reevaluate how management of people should be done, his character was assertive but compassionate. High morals coupled with grounded reason. I also want to mention his hilarious role as bullock in American dad, the way he takes a role both so seriously and ridiculously makes him so fantastically hilarious. I love him and will miss him.
[maxcan]: One thing that doesn't get enough attention - when COVID lockdowns started, he just made reels reading Shakespeare's sonnets. It was such a lovely break from everything else on social media. Really loved those.
[GodelNumbering]: "It is possible to commit no mistakes and still lose. That is not a weakness, that is life." - Jean-Luc Picard Good startup advice. Good life advice.
[tempestn]: Like everyone else here, Patrick Stewart as Captain Picard was a foundational part of my childhood. As it happens, I'm currently on bag 27/30 of the Lego Enterprise D right now. I've actually had the thought a couple of times over the past few years that this day had to be coming, and that it'd likely hit me hard when it did, especially after my dad's passing in 2024. (They're not particularly alike, aside from the hairline, but there's still a strong associ...
[madrox]: Patrick Stewart's life had the odds against in him in many ways. Abusive father suffering from PTSD. A bald british man in Hollywood invites typecasting. Constant imposter syndrome from growing up poor while entering a profession where many came from more educated or affluent backgrounds. Shakespeare effectively changed his life, and it's incredible to think that you could map everything he did and his success to his love of the bard. If I had to take a lesson from his life, it'...
Timothy B. Lee：Understanding AI（RSS） · 10-29 19:10 UTC
该文解读 Anthropic 的 S-1 文件，指出其财务表现强劲，但增长放缓存在隐忧。
AI 判断：标题初筛（未依据正文）：标题显示这是对 Anthropic S-1 财务文件的解读，属于融资与财务分析，并非普通人可直接使用的 AI 功能或教程。
展开原始摘录
作者通读 Anthropic 公开版 S-1 文件，认为公司财务状况好于预期：2026 年第二季度收入从一季度的 47 亿美元增至 116 亿美元，并录得首个运营利润，截至 6 月底现金储备仅次于 Google 和 Berkshire Hathaway。
Granola Adds Dictation Feature to Meeting Notetaker App, Rivaling Wispr
Future Tools · 10-29 18:00 UTC
Granola 在 Mac 端推出听写功能，按下 Fn 键即可在各应用中使用并整理口语文本，iOS 和 Windows 版即将推出。
AI 判断：普通人可亲自演示的跨应用语音输入与文本整理功能，适合办公场景，但需注意其目前仅限 Mac 且实际效果尚未验证。
展开原始摘录
Tools & Products · 5h ago Granola Adds Dictation Feature to Meeting Notetaker App, Rivaling Wispr Granola, the AI meeting notetaker app valued at $1.5 billion, has launched a dictation feature for Mac, directly competing with Wispr, a $2 billion dictation-focused startup that recently added its own meeting notetaker. Granola's tool activates via the "Fn" key, cleans up spoken text, and works across all apps for notes, emails, and prompts. iOS and Windows support is coming soon. The moves reflect a broader trend, with Fireflies, Speechify, and Monologue also converging into multifeature voice AI apps.
Alleged Chinese spy had 'close relationship' with Nvidia boss Jensen Huang
mgh2 · 10-29 17:43 UTC
有报道称一名被指为中国间谍的人士与英伟达CEO黄仁勋关系密切，属于地缘政治与安全类新闻，与AI产品使用无关。
AI 判断：这是未经独立核实的安全指控与政治新闻，既不涉及普通人可用的AI功能，也缺乏可演示画面，不适合作为AI教程或使用解读。
展开原始摘录
https://www.nytimes.com/2026/10/29/us/politics/nvidia-china-...
,
https://archive.ph/K7bih
--- Top Comments ---
[janalsncm]: A couple of questions after reading this article. The Post is not exactly top notch journalism. First, does her activity actually fit the definition of “espionage”? Espionage is the act of secretly gathering confidential, classified, or proprietary information without the owner's permission The most important question is what she actually sent and how secret it was. Most companies have levels of secrecy ranging from non public to highly sensitive. For example, schematics of their chip ...
[zetazzed]: Super weird that the only relationship documented in this article is two "dinners with AI executives in Asia." Does that count as a close relationship? I get invited to AI dinners in the Bay Area nonstop, and I understand they are always transactional (people wanting to sell me stuff), not actual relations with people who just happen to enjoy my presence...
[markus_zhang]: I doubt Jensen is completely out of the scheme TBH.
[motbus3]: I feel it is a bit naive to think Jansen would be a spy. He would hire all the spies and be the boss of both sides. It is not like he has to choose it
[Rapzid]: Please can we get this fixed to point to the NY Times article; they broke this story no? Also, New York Post? Let's do better.
Building a multiplayer game world with autonomous Fable-class models
Simon Willison · 10-29 17:00 UTC
文章标题称，作者使用自主的 Fable 级模型搭建了一个多人游戏世界。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是面向开发者的游戏世界搭建实验，普通人难以直接使用或复现。
展开原始摘录
How I Built the Most AI-Pilled Company in the World
Don't be put off by the clickbait headline, this piece by Jacob Sansbury describing how they are building
Spawn
is a fascinating exploration of the bleeding edge of what's possible with agentic engineering against frontier models.
Spawn is a massively multiplayer gaming environment where players can use prompts (to the in-game agent, Savi) to modify the game world.
I fired it up on my phone for a few minutes and was able to prompt a castle into existence, then a dragon, then ride the dragon to fly around that castle.
By Jacob's telling, Spawn is now being developed autonomously by a swarm of Fable-powered agents costing over $3 million in tokens
per month
("we spend 30x payroll on AI tokens").
Our agents change the live code directly and ship to production hourly, so they read every error, bug report and metric themselves. About 60% of our work starts from a production error, a Savi bug report (Savi is the AI you build games with in Spawn), something odd in our logs, or a feature request.
Savi files bugs herself. When a user hits one, or Savi hits one in the product, she files it. It goes straight to an agent that decides whether it's a real bug to fix, something we won't fix, or intentional design that needs a decision. Usually it becomes a change, delivered to that user within an hour, with a message in the same chat.
This is
continuous deployment
turned up to 11:
A loop this tight only works if it never stops, so we treat an hour without a ship as a SEV-0, our most severe kind of incident. Unshipped changes pile up, and every problem already in production lasts longer. The feedback slows down too. An agent can only keep a few hours of work in its head, so shipping faster than that actually helps it stay coherent. [...]
We can roll back a broken change in 30 seconds, exactly where it's broken and nowhere else. When a monitor or a log alert fires, it starts or wakes an agent that owns the incident, and we break things on purpose to test that those recovery paths work.
My eyebrows raised at least a dozen times reading this. I have trouble imagining this style of development working on software with higher stakes for when something goes wrong, but for a game this appears to be working surprisingly well.
Jacob says that Fable 5 (and then 5.1) was the model that made this level of autonomy possible. Opus 5.5 doesn't quite hack it for them, and when they tried GPT-6 Astra to apply fixes from a security review everything stopped working:
Every feature had been disabled and marked to be turned on again once the security fixes were done. Astra had gone through our whole product and made the insecure things secure by turning them off.
These kinds of extreme experiments are why I'm
excited about game development
as a substrate for exploring the edges of what today's coding agents can do.
Tags:
game-design
,
ai
,
generative-ai
,
llms
,
claude
,
coding-agents
,
agentic-engineering
,
claude-mythos-fable
Mistral Large 4 tops open models outside China with 63% cost cut and stronger coding
Future Tools · 10-29 17:00 UTC
Mistral 以自定义许可开源万亿参数模型 Large 4（Le Chonk），在 Artificial Analysis 智能指数上成为中国以外最强开源权重模型，单任务成本从5.54美元降至2.06美元。
AI 判断：这是面向开发者的开源模型发布，仅有跑分和API单任务成本，普通用户没有可直接使用的功能，不适合大众教程或效果演示。
展开原始摘录
Model Release · 6h ago Mistral Large 4 tops open models outside China with 63% cost cut and stronger coding Mistral has released open weights for its trillion-parameter Large 4 model, nicknamed Le Chonk, under a custom license. Independent evaluator Artificial Analysis scores it 26 points on Intelligence Index v5, making it the top open-weight model developed outside China, though it trails DeepSeek and GLM-5.3 globally. Cost per task dropped 63% from $5.54 to $2.06. On the Coding Agent Index, ML4 ranks second among open-weight models, using 47% fewer tokens than leader GLM-5.3. The license requires a separate commercial agreement for model-as-a-service operators.
OpenAI Updates Model Spec With 5 Core Values to Guide More Capable AI Systems
Future Tools · 10-29 17:00 UTC
OpenAI 更新了模型规范（Model Spec），围绕忠于人类意图、减少伤害、诚实透明、人类自主性与关怀这五项核心价值观重组行为准则，目的是让模型在规则未明确覆盖的情境下也能做出良好判断。
AI 判断：这是模型行为准则层面的政策性文档更新，没有普通人能直接上手使用或演示的功能，实用性和画面感都较弱。
展开原始摘录
Research · 6h ago OpenAI Updates Model Spec With 5 Core Values to Guide More Capable AI Systems OpenAI has updated its Model Spec, reorganizing behavioral guidance around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The update aims to help models exercise good judgment in situations not explicitly covered by instructions, as AI systems take increasingly consequential actions. New sections address context, oversight, and agency, while detailed style guidance has been retired. The spec also shifts from calling models "the assistant" to "GPT" to reflect broader deployment settings.
X：腾讯混元 (@TencentHunyuan) · 10-29 17:00 UTC
腾讯混元发布 Hy-Nomas 技术报告，并开源了名为 Hy-Nomas-Bench 的基准测试。
AI 判断：标题初筛（未依据正文）：仅从标题看是该团队的技术报告和基准测试开源，未提及普通用户可直接使用的功能或产品变化，属开发者/研究向内容。
展开原始摘录
腾讯混元发布 Hy-Nomas 开源交互式视频世界模型的技术报告（arXiv:2610.36729），并开源 Hy-Nomas-Bench 数据集用于可复现研究。官方还用 Hy-Nomas 重现 Genie 4 示例并放出对比视频，演示无需排队即可试用，地址为 https://hy-nomas.hy.tencent.com/。
OpenAI News · 10-29 17:00 UTC · 同链接合并评分
OpenAI 更新了模型规范（Model Spec），围绕忠于人类意图、减少伤害、诚实透明、人类自主性与关怀这五项核心价值观重组行为准则，目的是让模型在规则未明确覆盖的情境下也能做出良好判断。
AI 判断：这是模型行为准则层面的政策性文档更新，没有普通人能直接上手使用或演示的功能，实用性和画面感都较弱。
展开原始摘录
OpenAI’s updated Model Spec outlines five core values to guide more capable AI toward sound judgment, human agency, and faithfulness to human intent.
OpenAI：官网动态（RSS · 排除企业/客户案例） · 10-29 17:00 UTC · 同链接合并评分
OpenAI 更新了模型规范（Model Spec），围绕忠于人类意图、减少伤害、诚实透明、人类自主性与关怀这五项核心价值观重组行为准则，目的是让模型在规则未明确覆盖的情境下也能做出良好判断。
AI 判断：这是模型行为准则层面的政策性文档更新，没有普通人能直接上手使用或演示的功能，实用性和画面感都较弱。
展开原始摘录
OpenAI 更新其公开的 Model Spec，将指导原则重组为五项核心价值观：忠实于人类意图、最小化伤害、诚实透明、促进人类自主和关怀。更新新增了关于上下文理解、监督和自主性的指导，调整了部分行为默认值，退出了大部分详细风格指南，并将模型称呼从 the assistant 改为 GPT。
Codex 宣布即日起 Pro 及 Business Premium 各档位用量上限翻倍
X：Tibo (@thsottiaux) · 10-29 16:56 UTC
X 用户 Tibo 称 Codex 即日起把 Pro 和 Business Premium 各档位的用量上限提高一倍。
AI 判断：标题初筛（未依据正文）：这是仅面向编程开发者的订阅额度调整，普通用户无法实际使用或演示，属于标题层面的技术消息。
展开原始摘录
Tibo 宣布即日起将 Pro 100/200/500 档位与 Business Premium 的所有用量上限翻倍。Pro 200 的上限不会下调，Work 与 Codex 用量保持不变；dots 的扩展用量上限继续提供，且仍与默认用量分开计算，后续还会通过更好的模型和效率改进提升实际可用量。
X：Gemini (@GeminiApp) · 10-29 16:36 UTC
Gemini 新增可直接生成、预览和编辑文档、PDF、演示文稿及表格的功能。
AI 判断：标题初筛（未依据正文）：标题显示这是面向日常办公文档的实用新功能，适合演示生成与编辑效果，但具体可用范围和限制尚未披露。
展开原始摘录
Google Gemini 宣布可直接在应用内生成、预览和编辑文档、PDF、演示文稿、表格等多种文件。支持把手写笔记转成 PDF 学习指南、手动调整内容、高亮部分让 Gemini 修改文字或视觉元素，并可继续追问。
Google Official Blog · 10-29 16:15 UTC · 同链接合并评分
Gemini 应用更新后，用户可直接在聊天中生成、预览并编辑文档、PDF、演示文稿、电子表格及代码等文件，其中 PDF 预览与编辑支持网页和移动端，部分文件编辑仅限网页端。
AI 判断：这是普通人可直接上手的办公文件处理新功能，演示画面清晰且时效性强。
展开原始摘录
Generate, edit, and preview PDFs or Docs directly in Gemini. Use the new full-screen editor to refine your files today. See what's new in Gemini.
Gemini Notebook makes hundreds of textbooks available to help students learn even more.
Google Official Blog · 10-29 16:00 UTC
Gemini Notebook 宣布提供数百本教科书资源，以帮助学生学习更多内容。
AI 判断：标题初筛（未依据正文）：标题显示这是面向学生学习的实用功能更新，具备演示潜力，但仅凭标题无法确认开放地区与具体使用限制。
展开原始摘录
Add textbooks from OpenStax and S Chand to Google Gemini Notebook. Generate custom study guides and quizzes now. See the available catalog.
H.264 Is Magic (2016)
ea016 · 10-29 15:53 UTC
一篇2016年的文章解释H.264视频压缩为何高效。
AI 判断：这是旧技术科普，与普通人的AI新功能或实用场景没有直接关系。
展开原始摘录
--- Top Comments ---
[tgtweak]: Had the pleasure of working with Jason Garrett-Glaser (better known as DarkShikari) circa 2011 on a patch for x264 to better utilize more cores (slice-threaded lookahead). At the time he was working with Sony to optimize their PlayStation game streaming service, and this had the potential to lower encoding latency (critical to game streaming) even on same-core systems by spreading the frame over more cores to do more parallel work, and doing more intelligent scheduling and convergence after ...
[Dwedit]: One uncommonly used feature of H.264 is support for lossless video. It's a bit hard to use due to tooling (not easy to discover and select the feature), and is sometimes incompatible with web browsers, but you can get actual lossless video using a mainstream video codec. You can verify that it really is lossless, and doesn't have any conversion step where it's quantized to 8-bit YUV. "x264vfw" is one tool which supports this format, even inside of an AVI file. So o...
[pwython]: Ten years later, 2026 magic: AV1 https://engineering.fb.com/2026/06/22/video-engineering/adop... And soon AV2 https://av2.aomedia.org/
[HelloUsername]: Previous discussions: 4.nov.2016 https://news.ycombinator.com/item?id=12871403 219 comments 17.mar.2022 https://news.ycombinator.com/item?id=30710574 219 comments 24.may.2019 https://news.ycombinator.com/item?id=19997813 180 comments 14.jun.2024 https://news.ycombinator.com/item?id=40681306 205 comments
[Hasz]: I really like the idea that compression is intelligence, there are so many tricks an interesting ideas related to it. A bunch of them are described in the article, but you can imagine a long tail of many more inscrutable tricks chained together to get really incredible shrinkage in data. Hutter prize (I think mentioned in a 3b1b video somewhere else in this thread) is a great example. Given this was written before advent of LLM, I would love to see someone throw openAI/Anthropic levels o...
wondersky · 10-29 15:45 UTC
Grace Blackwell needs a new scheduler: 27% more usable GPUs from the same racks
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[jhgg]: This blog post spends a lot of time telling why gang scheduling is needed here (a fact obvious to anyone using the nvl72s), but doesn't actually go into detail about how their scheduler works. At work, we have built something very similar, a custom topology aware scheduler for workloads, and the most interesting things about such a project are entirely absent from this article, making it more of a fluff/marketing piece than an actually interesting technical blog in this area. Our ne...
[asaiacai]: There’s been some prior art in this space I feel that should be mentioned from k8s SIG batch in the Kueue project
https://kueue.sigs.k8s.io/docs/concepts/topology_aware_sched...
[timzaman]: Who has nvl72s that this isnt absolutely plainly obvious to - and has been for years. Not just that - topology has -always- been relevant to scheduling, and hpc schedulers have always been aware. What a weird blogpost.
[zan2215]: "Not all high-speed interconnects are created equal"
I think he's a red rising fan
[novalite]: This reads a bit disingenuous and click-baity. There's no "new" scheduler. Gang scheduling and rack-aware scheduling have been around since the days of HTCondor, and are put in practice at almost every large GPU deployment I've seen. The net new information added by this blog is close to zero, and feels like shilling for SkyPilot.
Zombie stats: Where the numbers AI quotes come from
mektrik · 10-29 15:06 UTC
这篇文章追踪 AI 回答中常被引用的“僵尸统计”数字的出处，并讨论这些数据的来源与可靠性。
AI 判断：选题贴近普通人使用 AI 时的信息甄别，但偏观点分析、缺少可演示效果，实用性和时效性一般。
展开原始摘录
--- Top Comments ---
[Isamu]: To me the dream of AI is: an agent can do the drudge work of tracking every statement down to the supporting reference, documentation and ultimately the work involved in verifying. It is NOT being used for this, instead we want faster “answers” that avoid due diligence. Same as always but lazier.
[toddmorey]: I can't decide if the report's writing ("One number, followed" & "Quoted by all five. Sourced by none.") is being lazy or ironic?
[MarkusQ]: And this doesn't even touch on the fact that a significant fraction of the published, peer-reviewed, authoritative sources ultimately turn out to be wrong[1]. Science is hard. Data collection and cleaning is hard. Mistakes are easy to make. [1] https://journals.plos.org/plosmedicine/article?id=10.1371/jo...
[falcor84]: A full 40% of AI stats actually have a specific study behind them? Knowing how the human baseline is that 73.6%* of all statistics are made up on the spot, this is really encouraging. *That particular stat was of course made up, but not by me.
[alberth]: I know Perplexity isn't often talked about anymore but their design of every sentence needing to be able to be referenced back to an actual source (like an academic peer reviewed paper) seems like a super power for trust.
LLM Voice
ibobev · 10-29 14:57 UTC
Martin Fowler 的短文讨论大语言模型语境下的“Voice”概念，偏开发与写作思考，而非可直接上手的工具教程。
AI 判断：来源内容信息有限且偏开发者讨论，缺少普通人可复现的操作或明确消费级新功能。
展开原始摘录
--- Top Comments ---
[BurritoKing]: As someone who is aggressively pro AI, I've also noticed how awful everything is to read lately. It's always so hard to put my finger on the exact thing, but it's a mixture of everything sounding identical and everything being a rambling unconcise mess. I find I routinely start reading interesting posts and just .. stop part way, abandoning it. A site I read stories on has been flooded with AI generated content, all of it terrible, and it just makes me retreat to things crea...
[PaulKeeble]: I just can't seem to stay interested and read LLM text. It just seems to lack going somewhere, its information density is often really low and its full of odd isms that I don't find very interesting to read. I don't have a good radar for detecting the various LLM particular speech patterns but everytime I have an issue with an article there is always people talking about how its clearly AI written. I find the same thing with written results out of AI "research", it is...
[Terr_]: All human communication is based on internally simulating the presumed mind that was/will-be at the other end. Even now, some part of your brain--however subconscious--is building a tiny sketch of me with assumed beliefs and motivations, and you're using it to predict and interpret the next words. But what happens when everything is persistently, subtly wrong ? When we cannot smoothly map the text to a simulation? An inner-sim keeps throwing errors and mis-predicting what comes nex...
[tptacek]: Broken record: LLMs are exceptionally useful writing critics if you know how to hold them correctly, but you can't use a single word they suggest to you --- the better the LLM's word choices seem, the more likely the LLM is infecting your voice. Better, by a lot, to prompt the LLM not to suggest any words, but just offer objective critiques.
[ceroxylon]: The voice is one thing, but I have encountered multiple 'tech support' teams and platforms that will respond to me in pure, unedited LLM output. I know this because the topic is usually an obscure issue, and I have already gone back-and-forth with multiple LLMs to see where their boundary of knowledge is on it, and it will often include hallucinations along the way (e.g. pointing to old/nonexistent settings pathways) So when the first reply from the official 'support'...
Wiz：Research（安全研究） · 10-29 14:36 UTC
Wiz 安全研究总结了智能体攻击的七种特征指纹，并梳理其发展趋势。
AI 判断：标题初筛（未依据正文）：这属于面向企业安全的智能体攻击研究，与普通人日常使用关联较弱，标题也未提供可演示的消费级功能。
展开原始摘录
Wiz 基于自身事件响应与公开披露案例，分析 AI 智能体参与真实入侵的现状与趋势，涉及 ARTEX、Hermes 等框架及韩国金融机构（约 67,000 人信息外泄）、泰国财政部、ConnectWise 供应链（尝试攻击超 2,250 个租户）等案例。
Unsloth 发布 Mistral Le Chonk 1-bit GGUF 量化版，可本地运行 1T 参数模型
X：Unsloth (@UnslothAI) · 10-29 14:27 UTC
仅据标题，Unsloth 宣布推出 Mistral Le Chonk 的 1-bit GGUF 量化版本，并称可在本地运行 1T 参数模型。
AI 判断：标题初筛（未依据正文）：这是面向开发者的量化与本地部署消息，普通用户难以直接使用，也缺少可演示的日常场景。
展开原始摘录
Unsloth 发布 Mistral Le Chonk（Mistral Large 4，1T 参数 MoE，52B 激活）的 Dynamic 1-bit GGUF，可在 Mac Studio 上本地运行，保留 91% top-1 准确率，体积约 233GB（较 1.1TB FP8 小 80%）。图为各量化档位给出硬件需求，1-bit 需 256-270GB 内存，最小 UD-IQ1_S 约占 233GB 磁盘，并支持文本与图像输入、1M 上下文和 160+ 语言。GGUF 与指南见 huggingface.co/unsloth/Mistral-Large-4-1T-A52B-GGUF 和 unsloth.ai/docs/models/mistral-large-4。
404 Media（RSS） · 10-29 14:21 UTC
404 Media 报道称，Proton 内部因 CEO Andy Yen 强推 AI 转型而出现恐惧文化争议。
AI 判断：标题初筛（未依据正文）：标题显示这是公司内部文化与 AI 转型调查，缺少普通人可直接使用的功能或可演示效果，实用性和受众相关性有限。
展开原始摘录
404 Media 采访九名现任与离职 Proton 员工并查阅内部文件，报道 CEO Andy Yen 强推 AI 转型引发的内部冲突。10 月公司 90% 代码由 AI 编写，多使用 Anthropic 的 Claude；Yen 称今年是最后一年招聘工程师，并宣布将以可能决定去留的重新面试评估员工 AI 使用情况。
Quoting Ben Hylak
Simon Willison · 10-29 14:06 UTC
Simon Willison 的这篇文章标题是引用 Ben Hylak 的话，但标题未提供具体引文内容。
AI 判断：标题初筛（未依据正文）：仅凭标题无法判断它是否涉及普通人可用的 AI 新功能或可演示案例。
展开原始摘录
every product i use should just be a headless mcp, that feeds into an agent, where i never have to think about anything underneath.
except for our product, of course.
in our product, customers love the ui and love us very much.
we're special that way.
—
Ben Hylak
Tags:
model-context-protocol
,
ai
AI is overwhelming and I’m doing it wrong
Ben's Bites · 10-29 14:03 UTC
Ben's Bites 作者在文中讲述自己被 AI 信息与工具淹没、觉得自己用错了方法。
AI 判断：标题初筛（未依据正文）：这是个人感受类观点文，仅凭标题无法确认是否有可演示的方法或具体事实支撑，实用性和新颖度都较有限。
展开原始摘录
Let’s think about what we’re doing here
纽约时报称中国情报人员瞄准英伟达 CEO 黄仁勋
The Decoder：AI News（RSS） · 10-29 13:42 UTC
《纽约时报》报道称，中国情报人员曾将英伟达CEO黄仁勋作为目标。
AI 判断：标题初筛（未依据正文）：标题显示这是高关注度人物的安全与地缘新闻，但缺乏普通人可直接使用的AI信息。
展开原始摘录
据《纽约时报》报道，美国情报机构认为英伟达 CEO 黄仁勋身边一名高管为中国提供情报，此人是新加坡数据中心运营商 Megaspeed 的 Alice Huang（黄乐）。
The Verge 解读 Anthropic S-1：约 8500 亿美元 compute 承诺押注递归自我改进
The Verge：AI（RSS） · 10-29 13:30 UTC
The Verge 解读 Anthropic 的 S-1 文件，称其约 8500 亿美元的算力承诺押注递归自我改进。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是融资与算力承诺类行业新闻，普通人的直接使用关联弱，也缺少可演示效果。
展开原始摘录
The Verge 作者分析 Anthropic S-1，认为其商业模式依赖高速增长：2026 上半年营收同比增长 14 倍至 $16.3 billion，净收入留存率 900%，但不可撤销采购承诺合计 $265.6 billion，加上未开始的租赁协议 $321.4 billion 及其他协议，未来 20 年承诺至少约 $850 billion。
Equipping healthcare professionals with practical AI skills
Google Official Blog · 10-29 13:00 UTC
该内容介绍面向医疗专业人员的实用 AI 技能培训，旨在提升其在医疗场景中的 AI 应用能力。
AI 判断：标题初筛（未依据正文）：标题显示这是面向医疗从业者的技能培训，普通大众相关性有限，且缺少可亲自演示的具体细节。
展开原始摘录
Introduction of the Google AI for Healthcare Professionals Certificate
DoorDash to acquire Instacart
WalterSobchak · 10-29 12:33 UTC
DoorDash 宣布将收购 Instacart，这是配送行业的一起重大并购，与 AI 新功能无直接关系。
AI 判断：这是非 AI 的商业并购新闻，缺少普通人可演示或使用的 AI 内容，仅有间接的市场影响。
展开原始摘录
--- Top Comments ---
[themightyquinn]: My initial take to this news was “what about anti trust?!” But I was surprised to learn that DD is essentially non existent in grocery delivery and instacart’s non grocery deliveries are powered by Uber Eats. For two massive brands they really don’t offer the same product.(edit: not effectively, at least) If anything maybe this helps Instacart/DD compete in grocery with the real giants in Walmart and Amazon. I wouldn’t be mad to see DOJ block this but I was just surprised to learn about ...
[zach]: Seems like it should be “DoorDash (YC S13) to Acquire Instacart (YC S12)”, if only to emphasize that this is by far the largest merger of YC companies.
[ashu1461]: Doordash as an app or a business model always feel scammy and full of dark patterns. With this merger I have a feeling that things are only going to go worse.
[scoofy]: Anti-trust legislation is all well and good, but worthless when politicians are corrupt.
[thisisauserid]: Perfectly reasonable to continue ordering private taxis for your burrito. Where do I invest?
rokostik · 10-29 12:19 UTC
If IPv4 were land
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[shubhamjain]: India: 47M China: 358M Both countries have roughly the same population. The IPv4 problem has been swept under the rug with CGNAT and a normal person doesn't even notice it. But it sucks. Forget having a dedicated IP, even on a broadband connection, you don't get a public IP that's truly yours. According to IPInfo[1], there are around 1,000 users sharing my IP. I get flagged for malicious activity, and online gaming is impossible without a Tailscale relay. Of course, much of thi...
[dzonga]: at a certain point in time there has to be a forcing function or event that causes people to migrate from ipv4 to ipv6.
[BirdieNZ]: From a Georgist perspective, IPv4 is actually a form of land (it's a finite "natural resource", although that's not quite proper), and you could make a case for "land rent" being taxed on the addresses. This is the same as radio broadcast frequencies, web addresses, actual land, un-mined gold/oil/coal/whatever, and other finite natural resources that create monopolistic tendencies.
[10111001]: The only thing that surprised me was that Canada was so small compared to other Western countries. But I suppose that might be an artifact. The map shows IPv4 proportional to land area, but a "fair" allocation would have IPv4 proportionate to population, not land area, and I'd assume Canada is (overall) less densely populated than most other developed countries.
[andix]: IPv6 got some real traction in the last decade. From around 10% adoption up to 50%. It's not dead, like many are claiming. It works fine, just keep implementing it.
What is turn taking in conversational AI: A guide to natural flow
ElevenLabs Blog · 10-29 12:00 UTC
该文介绍对话式 AI 中的话轮转换（轮流发言）机制，以及如何让对话更自然流畅。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是面向对话式 AI 设计的概念指南，偏技术解释，缺少普通人可直接上手的演示或新功能。
展开原始摘录
Turn taking decides when a voice agent speaks, pauses, or listens. Learn how it works, how barge-in is handled, and how to measure and tune it.
How we use ElevenAgents to save 1,200 hours a month in daily operations
ElevenLabs Blog · 10-29 12:00 UTC
ElevenLabs 的一篇博客介绍其团队如何用 ElevenAgents 在日常运营中每月节省 1,200 小时。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是企业内部运营自动化案例，对普通人的直接相关性和可演示性有限，且缺少使用限制与可复现细节。
展开原始摘录
How we use ElevenAgents to save 1,200 hours a month in daily operations across GTM, product, support, and engineering
How to use Gemini to update your resume
Google Official Blog · 10-29 11:33 UTC
介绍如何借助 Gemini 来更新和修改个人简历。
AI 判断：标题初筛（未依据正文）：标题显示这是面向求职者的简历修改教程，实用性强，但新颖度和普通人的日常相关性有限，具体效果需看正文。
展开原始摘录
Turn raw notes into high-impact, ATS-friendly resume bullets using Gemini. Follow these four steps to polish your professional profile. Get the details.
Chained PHP2Shell exploit roots 25 servers, including US Government servers
882542F3884314B · 10-29 10:42 UTC
一篇文章称，攻击者利用一条串联的 PHP2Shell 漏洞链控制了 25 台服务器，其中包含美国政府机构的服务器。
AI 判断：这是纯技术安全事件，与 AI 及普通人的日常使用无关，缺乏可演示的消费级价值。
展开原始摘录
--- Top Comments ---
[jakobdabo]: > the attacker chained in a second exploit: a Linux kernel vulnerability in the XFRM ESP-in-TCP subsystem Linux seems to have a huge attack surface that doesn't necessarily need to be there for every user/server. E.g. this came out today: CVE-2026-98432: Linux TCP Use-After-Free Risk and Patch Validation [1]. I just disabled mptcp on my systems since I don't use it. Why was it enabled by default? Can I disable this XFRM thing too? [1] https://www.penligent.ai&#x...
[genxy]: I thought we were smug about how hard it was to make a working exploit for this?
[athrowaway3z]: > A1 — Memory Exhaustion (~383MB): I'm actually somewhat surprised at the clumsiness here. I'm a complete amateur that had DeepSeek build a PHP2Shell exploit, and once the stealmybitcoin site went up and set an upload limit, i had DeepSeek try to shrink/amplify it. DeepSeek couldn't find anything but a 2x reduction, but reading the report of how it was done - https://github.com/KillaBoi/StealMyBitcoin-PoC - I was somewhat surprised it didn't f...
[colemannugent]: Reading this I'm almost certain the firewall vendor is Palo Alto Networks. They included a PAN-OS setting verbatim and the SELinux config matches my experience.
[xxmarkuski]: Dream is btw founded and led by Sebastian Kurz, who was head of the Austrian government. He had to step down because of a corruption scandal.
Google 三季度搜索增速不及预期，投资者担忧 AI 搜索变现拖累核心业务
The Decoder：AI News（RSS） · 10-29 10:16 UTC
谷歌三季度搜索业务增速低于预期，部分投资者担心AI搜索的商业化会拖累公司核心业务。
AI 判断：标题初筛（未依据正文）：标题仅涉及财报与投资者担忧，缺少可演示的普通人AI用法，实用性和画面感较弱。
展开原始摘录
Google 三季度搜索收入增长约 12%，低于分析师预期的 15%，市值一度缩水最多 10%。高管 Philipp Schindler 承认 AI mode 查询的强劲增长因变现率较低产生了影响，同时公司称搜索查询年增速达五年来最高，并计划在美国及国际市场扩大聊天机器人广告项目。
Google’s Gemini 4 Carbon tops the charts (for now)
The Rundown AI · 10-29 10:00 UTC
标题称谷歌 Gemini 4 Carbon 在榜单中排名第一，但未说明具体用途或普通用户可用性。
AI 判断：标题初筛（未依据正文）：仅凭标题只能判断为榜单/模型表现类信息，缺乏面向普通人的实用场景，按技术类谨慎评分。
展开原始摘录
PLUS: Turn a prompt into a poster with Canva Craft
How Gemini is helping students and teachers in the classroom
Google Official Blog · 10-29 09:00 UTC
标题介绍 Gemini 如何帮助学生和教师在课堂中使用。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是 Gemini 在教育场景的应用与影响案例，适合讨论 AI 进课堂，但缺少可演示的具体功能细节。
展开原始摘录
A teacher crouching next to a student seated at a desk as they talk. In front of the student are papers and a laptop.
How we detect frontier model misalignment
OpenAI News · 10-29 09:00 UTC
OpenAI 发布文章，介绍其检测前沿模型对齐失准的内部方法（仅据标题判断）。
AI 判断：标题初筛（未依据正文）：这是模型安全与对齐的内部检测方法说明，属于工程与研究流程内容，普通人既用不上也难以直观演示（基于标题判断）。
展开原始摘录
OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.
OpenAI：官网动态（RSS · 排除企业/客户案例） · 10-29 09:00 UTC · 同链接合并评分
OpenAI 发布文章，介绍其检测前沿模型对齐失准的内部方法（仅据标题判断）。
AI 判断：标题初筛（未依据正文）：这是模型安全与对齐的内部检测方法说明，属于工程与研究流程内容，普通人既用不上也难以直观演示（基于标题判断）。
展开原始摘录
OpenAI 发布长文，详解其针对前沿模型失准行为的多层监测系统，包括激活探针、基于模型的审查、调查智能体和自动化调查四层架构，并对 GPT‑6 Astra 及以上模型的全部 RL 运行、评估和内部 Codex 流量进行监测。
AIBase 中文 · 10-29 08:26 UTC
Anthropic 在 Claude Code v2.1.311 中新增环境变量 CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS，允许开发者彻底关闭删除文件等高危操作的最后一道权限确认。
AI 判断：这是面向开发者的命令行权限开关，普通用户既用不上也难演示，只能作为技术圈安全话题小范围讨论。
展开原始摘录
AI资讯 AI新闻资讯 正文 AI删文件终于可以不用再问了！Claude Code上线全新环境变量，赋予开发者绝对放行权限 发布于AI新闻资讯 发布时间 : 2026年10月29号 16:26 阅读 : 1 分钟 10月29日消息，据科技媒体 The Register 报道，Anthropic 近期在推出的 Claude Code v2.1.311版本中，悄然加入了一项全新的环境变量。通过启用该设置，用户可以让 AI 智能体在执行删除文件等高危操作时，不再逐次征求用户的批准。 这一全新的环境变量全称为 CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS，直译过来即为“更加危险地跳过权限检查”。 Anthropic 的技术人员 Daisy Hollman 在随后的回应中解释了该功能的由来。此前版本中内置的 --dangerously-skip-permissions（危险地跳过权限检查）选项虽然听起来权限极大，但实际上并未彻底放行所有高危操作，系统底层仍然会强制拦截部分明显的危险指令，例如递归强制删除用户主目录内容的 rm -rf ~ 等操作。而此次新增的极端环境变量，则是专门为了满足部分核心开发者的 极高 自主性需求，允许其彻底关闭这道最后的保护防线。 针对该变量过于直白且略带调侃意味的命名，以及其未被列入对应版本的官方更新日志所引发的种种质疑，Hollman 随后在社交平台上公开确认:“这不是玩笑。” 不过，需要指出的是，该功能并不会默认对所有用户开启，必须由开发者主动在环境中进行配置方可生效，同时组织内部管理员此前施加的相关安全策略和限制也将继续发挥作用。对于追求 极致 效率、希望赋予 AI 智能体完全自主操作权限的 资深 开发者而言，这一更新无疑大幅减少了繁琐的弹窗确认流程;但从系统安全的角度来看，如何在完全放权与防范误操作之间找到平衡，也将成为开发者在实际使用中需要认真权衡的新课题。 相关推荐 专为端到端大模型打造：新芯航途发布X7 Ultra智驾芯片，中国首发量产Armv9 车规级CPU 10月29日，XHEART新芯航途发布全新一代智驾大模型芯片X7Ultra。面对智驾从多个小模型向“端到端大模型”演进的趋势，其秉持“大模型定义芯片”理念，从真实量产模型底层需求出发，倒推重设计算架构、存储方案及数据流转方式，并为未来模型演进预留空间。文章还涉及核心算力与架构设计。 2026年10月29号 17:58 148.9k 百度确认吴甜离职，文心Unlimited OCR2开源实现任意页数解析 传闻百度副总裁、百度研究院副院长吴甜离职。百度公关经理回应称，吴甜已结束在百度任职，但百度技术工作持续推进。百度还称，10月27日发布并开源文心Unlimited OCR2，全球首次实现任意页数文档解析，OCR上限从数十页推向“无限”，达最先进水平，实测可一次解析《哈利·波特》全系列。 2026年10月29号 17:35 161.1k Anthropic 递给企业一把"监工钥匙"：Claude 能看清每个员工的 AI 开销 Anthropic 10月29日推出Claude智能报告，向Enterprise套餐开放。管理员可在组织设置一键开启，生成图表，查看团队用AI做了什么、产出与花费；可按团队、最长28天区间和指定Claude产品（对话、Claude Code、Cowork等）统计。在“AI会否替老板决定谁走人”的疑虑中，该工具为管理者提供了更清晰的AI使用管控手段。 2026年10月29号 16:50 159.3k GitHub发布Copilot Fusion，多模型自动编排提升编码效率与性价比 GitHub在Universe 2026发布Copilot Fusion智能编排层，可自动路由并组合OpenAI、Anthropic、Google等模型完成编码任务，开发者无需手动选模型。它整合既有自动模型选择技术与9月研究预览的Project HydraFusion，开发者在模型选择器启用后，系统按任务复杂度自动采用单一模型、级联或评审等执行模式。 2026年10月29号 16:15 168.5k 大晓机器人推出 ACE-Brain-0.6 空间智能 API，七项评测均分反超 GPT-6 Astra 10月29日，大晓机器人发布具身空间智能接口ACE-Brain-0.6API，向开发者开放“看懂空间”能力。该接口支持三类核心功能：识别目标、在三维空间精准定位目标，以及判断人和物体的可互动区域，相当于为机器装上理解物理环境的“空间感知神经”。同时，模型通过扩容空间锚定训练数据变得更聪明。 2026年10月29号 16:14 167.6k 智启未来，您的人工智能解决方案智库 简体中文 © 2026 AIBase 备案号：闽ICP备08105208号-24
GitHub发布Copilot Fusion，多模型自动编排提升编码效率与性价比
AIBase 中文 · 10-29 08:15 UTC
GitHub 在 Universe 2026 大会上发布 Copilot Fusion 智能编排层，能按任务复杂度自动在 OpenAI、Anthropic、Google 等模型间选择单一、级联或评审模式，并设三档努力等级以控制成本。
AI 判断：暂无 AI 评分理由
展开原始摘录
AI资讯 AI新闻资讯 正文 GitHub发布Copilot Fusion，多模型自动编排提升编码效率与性价比 发布于AI新闻资讯 发布时间 : 2026年10月29号 16:15 阅读 : 1 分钟 GitHub在Universe2026大会上正式发布Copilot Fusion，这是一款智能编排层，可自动路由并组合OpenAI、Anthropic、Google等多家AI模型完成编码任务，无需开发者手动选择模型。该功能整合了既有自动模型选择技术与9月研究预览的Project HydraFusion多模型编排能力。 开发者只需在模型选择器中选用Copilot Fusion，系统将根据任务复杂度自动决定使用单一模型、级联（Cascade）或评审(Critique)等执行模式，并匹配Standard(侧重效率与成本)、High(增强推理)或Max(追求 最强 能力)三档努力等级，避免简单任务产生不必要开销。 在Artificial Analysis Coding Agent Index评测中，Copilot Fusion Max获得60分，仅比运行Opus5.5Max的Claude Code低2分，但平均API成本约为一半、Token消耗减半、任务完成速度快约40%。该功能已在所有GitHub Copilot计划中全面可用，覆盖Visual Studio Code、CLI、应用及JetBrains IDE，不额外收取编排费用，仅按实际调用模型计费，并提供至2026年11月30日的30%模型费用折扣。企业管理员可控制可用模型并设为默认体验。此外，系统还支持实验性本地模型路由，例如在Windows设备上调用微软MAI Code1.1Flash。 微软同步推出由GitHub团队开发的Copilot harness，已赋能包括Copilot Studio、VS Code在内的多款产品，并通过Microsoft Foundry向开发者开放。多模型智能编排正成为AI编程工具的重要演进方向，有望在性能与成本之间实现更优平衡。 相关推荐 专为端到端大模型打造：新芯航途发布X7 Ultra智驾芯片，中国首发量产Armv9 车规级CPU 10月29日，XHEART新芯航途发布全新一代智驾大模型芯片X7Ultra。面对智驾从多个小模型向“端到端大模型”演进的趋势，其秉持“大模型定义芯片”理念，从真实量产模型底层需求出发，倒推重设计算架构、存储方案及数据流转方式，并为未来模型演进预留空间。文章还涉及核心算力与架构设计。 2026年10月29号 17:58 148.9k 百度确认吴甜离职，文心Unlimited OCR2开源实现任意页数解析 传闻百度副总裁、百度研究院副院长吴甜离职。百度公关经理回应称，吴甜已结束在百度任职，但百度技术工作持续推进。百度还称，10月27日发布并开源文心Unlimited OCR2，全球首次实现任意页数文档解析，OCR上限从数十页推向“无限”，达最先进水平，实测可一次解析《哈利·波特》全系列。 2026年10月29号 17:35 161.1k Anthropic 递给企业一把"监工钥匙"：Claude 能看清每个员工的 AI 开销 Anthropic 10月29日推出Claude智能报告，向Enterprise套餐开放。管理员可在组织设置一键开启，生成图表，查看团队用AI做了什么、产出与花费；可按团队、最长28天区间和指定Claude产品（对话、Claude Code、Cowork等）统计。在“AI会否替老板决定谁走人”的疑虑中，该工具为管理者提供了更清晰的AI使用管控手段。 2026年10月29号 16:50 159.3k AI删文件终于可以不用再问了！Claude Code上线全新环境变量，赋予开发者绝对放行权限 Anthropic在Claude Code v2.1.311版本中悄然加入环境变量CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS，允许AI智能体执行删除文件等高危操作时跳过用户逐次批准。技术人员Daisy Hollman对此作出回应。 2026年10月29号 16:26 157.4k 大晓机器人推出 ACE-Brain-0.6 空间智能 API，七项评测均分反超 GPT-6 Astra 10月29日，大晓机器人发布具身空间智能接口ACE-Brain-0.6API，向开发者开放“看懂空间”能力。该接口支持三类核心功能：识别目标、在三维空间精准定位目标，以及判断人和物体的可互动区域，相当于为机器装上理解物理环境的“空间感知神经”。同时，模型通过扩容空间锚定训练数据变得更聪明。 2026年10月29号 16:14 167.6k 智启未来，您的人工智能解决方案智库 简体中文 © 2026 AIBase 备案号：闽ICP备08105208号-24
爱范儿 · 10-29 07:03 UTC
一个不识字却很会说话的 AI。
AI 判断：标题初筛（未依据正文）：标题只给出反常识钩子，未说明具体功能、使用方式或对普通人的实际影响，判断基于标题。
展开原始摘录
机器为什么不能先说话，再认字？
#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
How Ropes & Gray is developing AI for transactional diligence
OpenAI News · 10-29 07:00 UTC
Ropes & Gray 正在开发用于交易尽职调查的 AI 应用。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是律所专业工作流案例，普通受众相关性和可演示性有限。
展开原始摘录
Ropes & Gray is developing AI for transactional diligence with OpenAI. Lawyers review grouped documents, summaries, and reports before advising clients.
Addy Osmani 谈你可能对 Claude 过度提示了
X：Addy Osmani (@addyosmani) · 10-29 06:13 UTC
Addy Osmani 在 X 上提出，用户可能对 Claude 进行了过度提示。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是提示方法观点，缺少可演示细节，实用性与新颖性有限。
展开原始摘录
Addy Osmani 撰文称新版 Claude 模型所需提示词比多数人习惯的更少，针对 Opus 5.5 建议用单一明确目标加验证条件让 Claude 自主工作，从 /effort medium 起步按需调整。
Ryo Lu: https://t.co/9JFKOz3lEv
Ryo Lu · 10-29 05:53 UTC
Ryo Lu 发布了一条仅含短链接的推文，未提供可核实的正文信息。
AI 判断：标题初筛（未依据正文）：标题仅为作者名和短链接，信息不透明，无法判断实用性或受众价值。
展开原始摘录
https://t.co/9JFKOz3lEv
Ryo Lu: augmenting human intellect when i started making software, i wasn’t dreaming of a factory. i...
Ryo Lu · 10-29 05:52 UTC
Ryo Lu 回顾自己做软件的初衷，谈的是增强人类智慧，而不是把软件开发做成一座流水线工厂。
AI 判断：标题初筛（未依据正文）：标题只是抽象的理念表达，没有可演示的功能或可核实的事实细节，对普通观众的实用性低。
展开原始摘录
augmenting human intellect
when i started making software, i wasn’t dreaming of a factory.
i loved that something on a screen could become an extension of my mind. a place to put a thought before i understood it. to move things around until new connections appeared. to make little systems that worked the way i worked, even if it made no sense to anyone else.
the computer wasn’t interesting because it could do work without me. it was interesting because of what i could do with it.
the early lineage of computing was built around this possibility. in 1962, Engelbart described augmenting human intellect as helping a person understand complex situations and find solutions that were previously out of reach. his starting point was the human being, not the autonomous machine.
he wasn’t describing a faster black box. he was looking at the whole relationship between a person, their tools, their language, their methods, and what they could learn to do with them. the capability belonged to that whole system.
somewhere along the way, we started telling a much smaller story about a much more powerful technology.
more automation. more agents. more software producing more software. more output from fewer people. more growth. more numbers to prove that something is happening.
and less attention to the person any of this is supposed to be for.
i don’t want to pretend computing had a pure innocent beginning. replacing human labor was one of its ambitions. even in 1960, Licklider distinguished human-computer symbiosis from systems where the remaining people mainly served the machine. he wanted computers to participate in the formation of ideas, not only execute problems already defined.
that distinction feels important – does the system help you discover what you mean? or does it only want you to submit a request, and wait outside while it works?
sometimes waiting is exactly what i want. please do the search. move the files. take away the repetitive work.
but not everything difficult is overhead.
sometimes moving the pieces around is how you understand them. sometimes making the thing changes what you wanted to make. sometimes you need to see the wrong version, touch it, take it apart, and discover a better question.
i want tools that leave room for that relationship.
in his 1981 paper on Smalltalk, Ingalls described a system meant to support everyone’s creative spirit. personal mastery mattered: an individual should be able to comprehend the system, rather than remain dependent on something permanently beyond their reach.
he also argued that computer models should fit the human mind, rather than demand the reverse.
not one perfect workflow for everyone. systems that different people can inhabit differently. someone who thinks through pictures. someone who needs to write. someone who understands a problem by building a simulation. someone whose way of organizing their life doesn’t resemble a SaaS dashboard.
a system you can reshape as your understanding changes.
this is what i want AI to make possible. not just pumping out software, but having a living relationship with it. being able to ask why something works this way. change one part without surrendering the rest. see what happened. undo it. learn enough to make the next change yourself.
the machine can become more capable without making the person smaller.
but that won’t happen just because we make the models better. we have to care about it. design for it. notice when we are making someone more dependent while calling it empowerment.
i don’t want to measure the future only by how much work happens without us.
i want to know what becomes possible for the person who couldn’t participate before.
what can they understand now? what can they express? what can they change without asking permission?
when i say augmenting human intellect, that is what i mean.
not a world that no longer needs us.
a world we have more ability to shape.
Partnering with Singapore to bring frontier AI to public health
Google Official Blog · 10-29 03:30 UTC
该文宣布与新加坡合作，把前沿AI应用于公共卫生领域。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是合作方向公告，缺少普通人可直接使用或可演示的具体功能，实用性有限。
展开原始摘录
Speakers on stage at the Google for Health Singapore Summit.
Peter Yang: The job that AI will never disrupt is selling your little hand-made trinkets at the local far...
Peter Yang · 10-29 03:12 UTC
Peter Yang 提出，AI 无法取代的工作之一，是在本地集市售卖自己制作的小手工艺品。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是关于 AI 与就业的观点式调侃，缺少可验证细节和可演示场景。
展开原始摘录
The job that AI will never disrupt is selling your little hand-made trinkets at the local farmer's market.
Farmer's markets are about to experience a renaissance, mark my words.
Thariq: I think the same is true of creative fields, like I see artists really worrying about AI but...
Thariq · 10-29 03:05 UTC
Thariq 认为创意领域也存在类似情况，并提到艺术家对 AI 的担忧，但标题未展开具体观点。
AI 判断：标题初筛（未依据正文）：仅据标题，这是一句未展开的个人观点，缺少可演示功能或事实细节，适合做轻量讨论而非实用教程。
展开原始摘录
I think the same is true of creative fields, like I see artists really worrying about AI
but I wish so badly that I could sketch animation ideas and character concepts for my agent to flesh out instead of trying to find examples
I don't have the language! https://t.co/Q6EPcpiXtu
Peter Yang: .@ChatGPT and browser use has been an absolute godsend for working through medical paperwork.
Peter Yang · 10-29 03:03 UTC
Peter Yang 表示，ChatGPT 的浏览器使用功能帮他处理医疗文书，体验非常好。
AI 判断：标题初筛（未依据正文）：标题只给出个人体验，缺少具体步骤和可验证细节，实用价值需补充演示。
展开原始摘录
.@ChatGPT and browser use has been an absolute godsend for working through medical paperwork.
Garry Tan: There’s a budget crush and high schoolers can’t take honors classes (they’re phasing out) but...
Garry Tan · 10-29 02:59 UTC
Garry Tan 指出预算紧缩导致高中荣誉课程被逐步取消，学生无法选修。
AI 判断：标题初筛（未依据正文）：标题仅涉及教育预算与课程取消，与 AI 无关，缺乏对普通 AI 用户的实用价值。
展开原始摘录
There’s a budget crush and high schoolers can’t take honors classes (they’re phasing out) but guess what new classes they are getting next https://t.co/HXjeqjtuLk
Nan Yu: HEY JOHN APPLE 5G ON MY MACBOOK SEND TWEET
Nan Yu · 10-29 02:55 UTC
这是一条含义不明的推文式短句，仅从标题看像是在说在 MacBook 上使用 Apple 5G 网络发推文。
AI 判断：标题初筛（未依据正文）：标题是碎片化的口语短句，缺乏可核实的信息和可演示内容，无法判断是否有普通人可用的 AI 价值（基于标题判断）。
展开原始摘录
HEY JOHN APPLE 5G ON MY MACBOOK SEND TWEET
Swyx: Gemini #1 on AAv5??? this is making me open up @antigravity for the first time in a long time...
Swyx · 10-29 01:58 UTC
仅据标题：Swyx 称 Gemini 在 AAv5 榜单中排名第一，并表示这让他很久以来首次打开 @antigravity。
AI 判断：标题初筛（未依据正文）：仅凭标题可见这是对跑分榜单和开发者工具的个人感叹，缺少可核实细节，普通用户难以实际使用或复现。
展开原始摘录
Gemini #1 on AAv5???
this is making me open up @antigravity for the first time in a long time. i’m getting an early review, will report back soon.
if google actually pulls this off it is officially comeback season with Meta, Grok and now GDM being cool again.
@saranormous talked about model neurodiversity last year. we have roughly 2x more frontier labs than when she was last on the aie stage. absolutely bizarre counter example to the bitter lesson.
Thibault Sottiaux: Day 24/ Text your dot just like anyone else. Makes it super easy to share/forward things, lea...
Thibault Sottiaux · 10-29 01:49 UTC
标题称第24天可以像给任何人发消息一样给 dot 发消息，并称这让分享和转发更简单。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是含义不清的功能预告，缺少可核实的产品细节和普通用户可用信息。
展开原始摘录
Day 24/ Text your dot just like anyone else. Makes it super easy to share/forward things, leave it voice notes or you can even go creative with "Hey Siri, text Iggy to <INSERT WILD REQUEST>" and sit back and relax. https://t.co/T4t4m1m6DV
Nikunj Kothari: Been sharing this feedback to a lot of seed founders on how to bridge the gap between their d...
Nikunj Kothari · 10-29 01:15 UTC
一位投资人在社交平台表示，自己正把一份关于如何弥合差距的反馈分享给许多种子轮创始人；标题未说明具体所指。
AI 判断：标题初筛（未依据正文）：仅凭标题可见这是面向种子轮创始人的创业/融资建议，与普通人可用的 AI 功能无关，且标题被截断、信息不足，难以核实。
展开原始摘录
Been sharing this feedback to a lot of seed founders on how to bridge the gap between their dream to what is often required at the A from someone who sits on the other side of the table..
Thanks @dflieb, @adityaag & a few seed founders for reviewing drafts of this! https://t.co/e6k8pyXGc5
Josh Woodward: @Strava You can turn it on here: https://t.co/wAEcM3IYP1
Josh Woodward · 10-29 00:34 UTC
Josh Woodward 回复 Strava 称可在此处开启某项功能。
AI 判断：标题初筛（未依据正文）：仅凭标题只见回复与链接，缺少可核实的 AI 功能或普通人可演示内容，难以构成有用选题。
展开原始摘录
@Strava You can turn it on here: https://t.co/wAEcM3IYP1
Matt Turck: This is why French founders are a special breed - to even start a company, they have to first...
Matt Turck · 10-29 00:07 UTC
Matt Turck 发帖称法国创始人是“特殊的物种”，并暗示他们在创业之前必须先跨过某道门槛，但标题本身没有交代具体内容。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是一条关于法国创业者文化的悬念式个人感慨，与 AI 新功能、工具或普通人可用性无关，且关键信息被省略，无法核实（基于标题判断）。
展开原始摘录
This is why French founders are a special breed - to even start a company, they have to first face this kind of feedback from their families and friends lol 😂 https://t.co/XU6m5JcHcO
爱范儿 · 10-29 00:00 UTC
一篇早报汇总声称触屏MacBook Pro发布、发光logo回归、Gemini 4 Carbon发布并超越Opus 5.5，以及马斯克招聘寿司厨师等消息。
AI 判断：标题初筛（未依据正文）：标题为多话题早报且包含难以核实的产品与模型说法，缺乏可演示的实用信息，仅凭标题不适合作为教程或可靠新闻。
展开原始摘录
· OpenAI拟采购GoogleTPU，已招聘工程师适配前沿模型
· 仰望行政旗舰定名「鼎曜」，计划 11 月初在中东亮相
· 宇树与红杉中国成立「杉宇·创计划」，投资机器人早期项目
期项目
#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
vLLM 发布 Mistral Large 4 Day 0 支持
vLLM 官方博客（RSS） · 10-29 00:00 UTC
vLLM 官方博客宣布，在 Mistral Large 4 发布当天即提供对该模型的 Day 0 支持。
AI 判断：标题初筛（未依据正文）：这是推理引擎对新模型的适配更新，属开发者工具链消息，普通用户既无法直接使用，也缺少可演示的实际收益（仅据标题判断）。
展开原始摘录
vLLM 团队宣布对 Mistral 新旗舰模型 Mistral Large 4 提供 Day 0 支持，该模型为 1T 参数、每 token 激活 49B 的原生多模态 MoE 模型，采用 MLA 稀疏注意力加 DSA top-k 索引器、跨层 KV 共享、NoPE、SSMax 和 LeRoPE 等新架构，并内置 DSpark 推测解码草稿器与 Pixtral 视觉编码器。
Mistral Large 4（Le Chonk）登陆 Hugging Face：1T 参数 MoE 多模态开源权重模型
Hugging Face 官方团队博客 · 10-29 00:00 UTC
标题显示，Mistral Large 4（Le Chonk）以 1T 参数 MoE 多模态开放权重模型的形式登陆 Hugging Face。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是面向开发者的超大模型发布，普通人缺少可直接使用的场景与演示。
展开原始摘录
Hugging Face 发布 Mistral Large 4（Le Chonk）的上线博客，模型为多模态多语言 MoE 架构，总参数约 1T、激活 49B、上下文窗口 1M token。
OpenRouter 教程：用 LLM 从 PDF 中抽取表格
OpenRouter：Announcements（RSS） · 10-29 00:00 UTC
OpenRouter 发布教程，介绍如何调用 LLM 从 PDF 文件中抽取表格数据。
AI 判断：标题初筛（未依据正文）：基于标题判断，这是偏开发者 API 的教程；虽然对应 PDF 表格提取需求，但普通用户难以直接上手，且缺少可验证的操作细节。
展开原始摘录
OpenRouter 发布教程，讲解如何把 PDF 以文件形式连同严格 JSON schema 发给能读文件的模型，让表格以带类型的行返回，括号亏损、千分位和脚注标记等清理都写进字段描述。
Thibault Sottiaux: Team is working on this but it's not rolling out just yet, sorry for any confusion. Plan is t...
Thibault Sottiaux · 10-28 23:35 UTC
Thibault Sottiaux 表示团队正在处理这项功能，但暂时还未上线，并为可能造成的困惑致歉。
AI 判断：标题初筛（未依据正文）：仅凭标题只能看出这是对尚未上线功能的简短说明，缺少普通用户可用的具体功能、演示或可核实细节。
展开原始摘录
Team is working on this but it's not rolling out just yet, sorry for any confusion. Plan is to bring together best of both Chat / Work capabilities together at some point this year, but we're going to proceed carefully here and take our time.
We'll have more to share soon. https://t.co/St57fti6qW
Quoting Ted Sanders
Simon Willison · 10-28 23:23 UTC
这是一篇引用 Ted Sanders 言论的内容。
AI 判断：标题初筛（未依据正文）：标题仅显示引用他人言论，缺少可核实的具体信息与普通人可用价值。
展开原始摘录
In general, LLM performance is incredibly multidimensional. There's no guarantee of a fixed ordering where the model that is best at math is also the model that's best at jokes. Despite that complexity, much of the variation in performance is explained by just a single variable: model size. In general, models that are bigger will (a) be more performant / reliable, (b) cost more per token, and (c) have better recall of esoteric facts. As you note, our models are ordered:
Luna < Terra < Sol < Astra
[
...
] These days, version number signifies a few things at once: the order of release, the relationship to other releases, and the magnitude of improvement. So something like 6.1 would indicate it was launched after 5.6, may share some similarities to other 6.1 models in terms of training data / personality (which is why it's not 5.7), and isn't a giant enough leap to merit the headlines and expectations that would come with a 7.
—
Ted Sanders
, explaining OpenAI's model naming conventions
Tags:
generative-ai
,
gpt
,
openai
,
ai
,
llms
,
ted-sanders
Nikunj Kothari: https://t.co/YR5yYDTHc9
Nikunj Kothari · 10-28 23:14 UTC
这是一条只含短链接、没有正文说明的推文，无法判断具体内容。
AI 判断：标题初筛（未依据正文）：仅有标题和链接，缺少可核实的信息与可演示内容，只能按不确定选题处理。
展开原始摘录
https://t.co/YR5yYDTHc9
Josh Woodward: New in Gemini: Connect your @Strava account to turn years of raw workout logs into your perso...
Josh Woodward · 10-28 23:04 UTC
Gemini 新增 Strava 账号连接功能，可将多年的原始训练记录转化为个性化内容。
AI 判断：标题初筛（未依据正文）：这是面向健身用户的消费级功能，可演示原始日志与个性化结果的对比，但标题未说明开放地区、价格与使用限制，需进一步核实。
展开原始摘录
New in Gemini: Connect your @Strava account to turn years of raw workout logs into your personalized coaching plan https://t.co/iQXRYjMzoX
