【文章标题】：[AINews] GPT-6 Astra：OpenAI史上最大规模的LLM发布
【文章正文】： The launch is barely 9 hours old, and with 36M views and 164K likes, already is OpenAI’s most successful launch since Sora and certainly GPT-4 or GPT-5 . 发布仅过去9小时，就已获得3600万浏览量和16.4万点赞，成为自Sora以来OpenAI最成功的发布，当然也超越了GPT-4或GPT-5。
You’ll recall we’ve previously observed that Anthropic tends to far outclass OpenAI in launch popularity. For the first time in their mutual history , OpenAI has turned the tables. 您可能记得我们之前观察到Anthropic在发布热度上往往远超OpenAI。但在双方历史上首次，OpenAI扭转了局面。
You can read our initial impressions here and we will update with more coverage soon, just stay subscribed. Overall a very welcome answer to Anthropic’s Fable and Opus progress. Your move, SpaceXAI and Google DeepMind. 您可以在此处阅读我们的初步印象，我们将很快更新更多报道，请保持关注。总体而言，这是对Anthropic的Fable和Opus进展的一个非常受欢迎的回应。现在轮到SpaceXAI和Google DeepMind行动了。
AI News for 9/2/2026-9/3/2026. We checked 12 subreddits, 544 Twitters and no further Discords. 2026年9月2日至3日的AI新闻。我们检查了12个Reddit子版块、544条Twitter推文，但没有更多的Discord消息。
AINews’ website lets you search all past issues. As a reminder, AINews is now a section of Latent Space . You can opt in/out of email frequencies! AINews的网站可让您搜索所有过去的期刊。提醒一下，AINews现在是Latent Space的一个栏目。您可以选择加入/退出电子邮件频率！
AI Twitter Recap AI Twitter回顾
OpenAI launched GPT-6 Astra as its new flagship model, but the rollout and the surrounding debate were almost as consequential as the model itself. OpenAI发布了GPT-6 Astra作为其新的旗舰模型，但推出过程和围绕它的辩论几乎与模型本身一样重要。
OpenAI officially announced Astra as “our most intelligent and aligned model yet,” positioning it around computer use, software engineering, math/science, polished office work, and cybersecurity via @OpenAI , @OpenAI , and @sama OpenAI官方宣布Astra是”我们迄今为止最智能且对齐的模型”，通过@OpenAI、@OpenAI和@sama将其定位为围绕计算机使用、软件工程、数学/科学、精细的办公室工作和网络安全。
The company said Astra was rolling out first to a limited set of organizations, then over days to ChatGPT Plus/Pro/Business/Enterprise, the API, and AWS, as noted by @OpenAI , @OpenAIDevs , and @thsottiaux 该公司表示，Astra将首先向有限的组织推出，然后在几天内向ChatGPT Plus/Pro/Business/Enterprise、API和AWS推出，如@OpenAI、@OpenAIDevs和@thsottiaux所述。
The launch itself was bumpy: users saw delays, a broken/late blog post, unclear access timing, and frustration that many influencers had early access while paying users did not, as reflected by @iScienceLuvr , @kimmonismus , @sama , @sama , @sama , @theo , and @t3dotcodes 发布本身并不顺利：用户经历了延迟、博客文章发布失败/延迟、访问时间不明确，以及许多影响者提前获得访问权限而付费用户没有的沮丧，如@iScienceLuvr、@kimmonismus、@sama、@sama、@sama、@theo和@t3dotcodes所反映。
OpenAI tried to compensate for delays by granting “banked resets” for each day paid ChatGPT users lacked Astra access, per @thsottiaux and @reach_vb 根据@thsottiaux和@reach_vb的说法，OpenAI试图通过为付费ChatGPT用户每天缺少Astra访问权限提供”存储重置”来补偿延迟。
OpenAI simultaneously released a system card / deployment safety material that drew unusually intense attention because it described both improved alignment and decreased chain-of-thought monitorability, highlighted by @scaling01 , @tomekkorbak , @MicahCarroll , and @kaicathyc OpenAI同时发布了一份系统卡/部署安全材料，引起了异常强烈的关注，因为它描述了改进的对齐性和降低的思维链可监控性，如@scaling01、@tomekkorbak、@MicahCarroll和@kaicathyc所强调。
Astra’s benchmark profile immediately triggered dispute: OpenAI and sympathetic testers described a step-change or “AGI-like” leap; independent aggregators and some researchers argued the gains were large but uneven, especially once cost and non-cherry-picked evals were considered, e.g. @ArtificialAnlys , @arcprize , @fchollet , @EpochAIResearch , @theo , and @abacaj Astra的基准测试立即引发了争议：OpenAI和支持的测试者描述了一个阶跃变化或”类似AGI”的飞跃；独立聚合者和一些研究人员认为收益很大但不均衡，特别是在考虑成本和非精选评估时，例如@ArtificialAnlys、@arcprize、@fchollet、@EpochAIResearch、@theo和@abacaj。
The strongest positive reactions centered on computer use, 3D generation/reconstruction, game-building, long-horizon knowledge work, and formal/scientific reasoning, from a mix of OpenAI staff, benchmark authors, partners, and early testers such as @markchen90 , @mckbrando , @Dimillian , @theo , @MattShumer_ , @skirano , @tomkrcha , @realYunfanYe , @nasqret , and @rileybrown 最强烈的积极反应集中在计算机使用、3D生成/重建、游戏构建、长期知识工作和正式/科学推理上，这些反应来自OpenAI员工、基准测试作者、合作伙伴和早期测试者的混合，如@markchen90、@mckbrando、@Dimillian、@theo、@MattShumer_、@skirano、@tomkrcha、@realYunfanYe、@nasqret和@rileybrown。
The strongest negative reactions centered on monitorability, evaluation-awareness, release governance, benchmark saturation, and the possibility that visible alignment gains are partly “papering over” specific failure modes rather than solving underlying goal misalignment, especially from @NeelNanda5 , @RyanGreenblatt , @RyanGreenblatt , @RyanGreenblatt , @scaling01 , and @teortaxesTex 最强烈的负面反应集中在可监控性、评估意识、发布治理、基准饱和以及可见的对齐收益可能部分”掩盖”特定故障模式而不是解决潜在目标错位的可能性上，特别是来自@NeelNanda5、@RyanGreenblatt、@RyanGreenblatt、@RyanGreenblatt、@scaling01和@teortaxesTex。
Official claims and concrete specs 官方声明和具体规格
OpenAI’s public positioning combined capability claims, benchmark claims, deployment claims, and product claims. OpenAI的公开定位结合了能力声明、基准测试声明、部署声明和产品声明。
Core announcement language: Astra is the “most intelligent and aligned model yet” and “Anything you can do on a computer, Astra can do for you. Fast.” via @OpenAI 核心公告语言：Astra是”迄今为止最智能且对齐的模型”，“任何你能在计算机上做的事情，Astra都能为你快速完成。“通过@OpenAI。
Model capabilities emphasized by OpenAI: OpenAI强调的模型能力：
state-of-the-art computer use and software engineering 最先进的计算机使用和软件工程
“new breakthroughs” in math and science 数学和科学的”新突破”
polished documents/spreadsheets/presentations following templates/style 遵循模板/风格的精细文档/电子表格/演示文稿
stronger cybersecurity capabilities with monitoring/safeguards 具有监控/保障的更强大的网络安全能力
via @reach_vb , @OpenAIDevs , @OpenAIDevs 通过@reach_vb、@OpenAIDevs、@OpenAIDevs。
Availability: 可用性：
limited org rollout first 首先向有限的组织推出
then Plus, Pro, Business, Enterprise 然后是Plus、Pro、Business、Enterprise
API and AWS over coming days API和AWS在未来几天内推出
via @OpenAI , @OpenAIDevs 通过@OpenAI、@OpenAIDevs。
Pricing: 定价：
standard: 50 / 1M output tokens 标准：每100万输入token 10美元，每100万输出token 50美元
fast: 100 / 1M output 快速：每100万输入token 20美元，每100万输出token 100美元
, for up to 2.5x speed 速度最高可达2.5倍
via @reach_vb 通过@reach_vb。
Product/runtime features announced alongside Astra: 与Astra一起宣布的产品/运行时功能：
Codex can ask questions while continuing independent work Codex可以在继续独立工作的同时提问
experimental context feature that lets Astra keep notes and search earlier context windows during long tasks 实验性上下文功能，让Astra在长时间任务中保留笔记并搜索早期的上下文窗口
Responses API additions: Responses API新增功能：
async function calling 异步函数调用
, mid-turn steering 中途转向
, and changing reasoning effort without breaking cache 在不破坏缓存的情况下改变推理努力
via @reach_vb , @nikunjhanda 通过@reach_vb、@nikunjhanda。
Claimed benchmark figures from OpenAI comms: OpenAI通讯中声称的基准测试数据：
99.9% on ARC-AGI-3 ARC-AGI-3上99.9%
98% on FrontierMath Tier 4 FrontierMath Tier 4上98%
100% on ExploitBench ExploitBench上100%
1.9x faster than GPT-5.6 Sol on Mind2Web 在Mind2Web上比GPT-5.6 Sol快1.9倍
with Codex harness improvements 通过Codex的改进
via @reach_vb , @sama 通过@reach_vb、@sama。
OpenAI also claimed Astra had “already helped so OpenAI还声称Astra已经帮助了…
🔗 知识库双向关联
- [[Raw_翻译_[AINews] Death of Params Z.ai CEO Jie Tang on GLM 5.3 and th|Death of Params Z.ai CEO Jie Tang on GLM 5.3 and th_全翻译]]
- [[Raw_翻译_[AINews] Memory prices up 500 in 12 months|Memory prices up 500 in 12 months_全翻译]]
- Weekly Dose of Optimism 205_简报