[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-microsoft-token-budget-gpt-5-6-default":3,"topics-all":41,"news-related-9e58d587-3c1b-44c5-ad36-daf23aeb42a2":60},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":27,"news_slug":34,"published_at":35,"created_at":36,"modified_at":37,"is_published":38,"publish_type":39,"image_url":14,"view_count":40},"9e58d587-3c1b-44c5-ad36-daf23aeb42a2","微软叫停 tokenmaxxing:GitHub Copilot 默认切回 GPT-5.6 Sol,Parikh 设 token 预算","一份微软内部薪酬表上的 AI 支出字段揭示 token 通胀:一名工程师 28 天花 2.8 万美元,公司整体中位数仅 300 美元,CoreAI 中位数 975 美元。执行副总裁 Jay Parikh 发邮件叫停,并把 Copilot 默认模型从 Claude 切回 GPT-5.6 Sol。","过去两年,大厂对工程师的态度一直是\"多用 AI、多烧 token\"。微软把这套口号第一次拉回了\"业务结果\"刻度。本月早些时候,执行副总裁、CoreAI 工程部门负责人 Jay Parikh 向全员发出内部邮件,要求工程师\"专注于业务成果,而非最大化 AI token 的使用量\",并明确要从 token 投资中获取更大价值。Times of India 在 8 月初转引了这份内部备忘(原始邮件由 404 Media 率先获得)。\n\n真正的成本动作不是邮件,而是把 GitHub Copilot 默认模型从 Claude 切回 OpenAI GPT-5.6 Sol。Times of India 援引 Parikh 邮件原文:\"tokenmaxxing is not what we are optimizing for\"。GPT-5.6 Sol 在 7 月推出,微软把数十万员工每天的 Copilot 默认流量全切到这一档,这是 OpenAI 与 Anthropic 收入分配的一次实际再平衡。\n\n## 导火索:一份 2.8 万美元的账单\n\nSolidot 把这场整顿的导火索拆得更清楚:一份新字段\"AI $ Usage Per Month\"挂在了内部员工薪酬表上。在逾 22.3 万名微软员工中,约 350 名美国员工自愿提交了当月 AI 工具支出,最触目惊心的数字来自 Customer and Partner Solutions 部门——一名工程师 28 天 AI 支出高达 2.8 万美元,公司整体中位数约为每 28 天 300 美元;部门差距明显:CoreAI 中位数 975 美元,Security 526 美元。Solidot 描述这种失控为\"打副本刷经验\":仪表板可见指标演变成排行榜,刷分就成目标。\n\n## 所以呢\n\n云时代有过一模一样的剧本:企业喊\"想开多少机器就开多少\",直到 AWS 账单冲到六位数才出来建仪表板。AI token 现在正走这条路。三个信号:可见指标必须搭配业务配对,否则必被刷成游戏;默认模型是最大成本杠杆,一次切换的收入影响远大于市场宣传;大厂 AI 优先战略也会撞上 unit economics 的墙。Times of India 援引数据,过去三年 per-token 价格跌约 98%,但企业 AI 账单反增 3 倍——亚马逊、Adobe、Atlassian、Citi、Meta 都比微软更早做 token 节流,Meta 还关停\"Claudeonomics\"内部排行榜。\n\n如果你的 Copilot 仪表板只显示 token 数、不显示业务结果,准备一份新的预算争议;如果默认模型是闭源贵模型,准备切换路径;\"AI 优先\"作为文化口号可以保留,但作为预算口径必须配套 FinOps 纪律。\n\n(来源:[Times of India](https:\u002F\u002Ftimesofindia.indiatimes.com\u002Ftechnology\u002Ftech-news\u002Fmicrosofts-top-ai-boss-has-a-message-for-engineers-using-github-copilot-your-token-spend-is-now-being-tracked\u002Farticleshow\u002F133046474.cms)、[404 Media](https:\u002F\u002Fwww.404media.co\u002Fmicrosoft-tells-engineers-tokenmaxxing-is-not-what-we-are-optimizing-for\u002F))","https:\u002F\u002Ftimesofindia.indiatimes.com\u002Ftechnology\u002Ftech-news\u002Fmicrosofts-top-ai-boss-has-a-message-for-engineers-using-github-copilot-your-token-spend-is-now-being-tracked\u002Farticleshow\u002F133046474.cms","7d7087f3-439e-462f-9848-52a3eaad838c",[11,15,18,21,24],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"id":19,"name":20,"slug":20,"description":14,"color":14},"baf131c1-687a-49f4-87f6-4dd87c1c692f","gpt",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":25,"name":26,"slug":26,"description":14,"color":14},"42e59a88-7795-47dc-a334-ef1e72c24347","openai",[28],{"id":29,"lang":30,"title":31,"summary":32,"content":33},"7412d8d5-68d9-4fb6-b2cd-2b7376ca9b9d","en","Microsoft ends tokenmaxxing: Copilot defaults to GPT-5.6 Sol","An AI spending column that surfaced on Microsoft's internal payroll revealed token inflation: one engineer spent $28,000 over 28 days while the company-wide median sat near $300 and CoreAI's median hit $975. EVP Jay Parikh sent a memo telling engineers to stop, and the company shifted Copilot's default model from Claude back to GPT-5.6 Sol.","## Background\n\nFor the past two years, the line pushed to engineers inside large AI labs has been simple: use more AI, burn more tokens. Microsoft is the first major player to walk that slogan back to a 'business outcomes' axis. Earlier this month, Executive VP of CoreAI Jay Parikh sent an internal memo asking engineers to focus on outcomes that move the needle for customers and the business, rather than maximizing AI token usage, and to extract greater value from the company's token investment. Times of India reported on the memo in early August (the underlying note was first obtained by 404 Media).\n\n## What actually moves the needle\n\nThe real cost action was not the memo itself but shifting GitHub Copilot's default model from Claude back to OpenAI's GPT-5.6 Sol. Times of India quoted Parikh's memo directly: 'tokenmaxxing is not what we are optimizing for.' GPT-5.6 Sol shipped in July, and Microsoft rerouted the default traffic of hundreds of thousands of employees on Copilot onto that tier. That is a real-world rebalancing of revenue allocation between OpenAI and Anthropic, not a symbolic gesture.\n\n## What triggered the cleanup\n\nSolidot laid out the trigger in more concrete terms: a new 'AI $ Usage Per Month' column appeared on the internal payroll spreadsheet. Among more than 223,000 Microsoft employees, roughly 350 US-based staff voluntarily reported their monthly AI tooling spend. The most striking number came from the Customer and Partner Solutions organization: one engineer logged $28,000 in AI costs over a 28-day period, while the company-wide median sat near $300 per 28 days. Department-level medians diverge sharply: CoreAI at $975, Security at $526. Solidot described the loss of control as a leaderboard dynamic: visible Copilot dashboards turn into a ranking, and once the ranking exists, climbing it becomes the goal.\n\n## So what\n\nThe cloud era already ran this script. Companies said 'spin up whatever you need' until AWS invoices hit six figures and someone finally built a dashboard. AI tokens are now on the same stretch of road. Three signals to remember: visible metrics must be paired with business attribution or they will be gamed; the default model is the largest cost lever, and one internal switch moves revenue more than any marketing campaign; and the AI-first strategy at hyperscalers is hitting the wall of unit economics. Times of India noted that per-token pricing has fallen roughly 98 percent over the past three years, yet enterprise AI bills have tripled, because agentic tools consume tokens at a rate that no autocomplete tool ever did. Amazon, Adobe, Atlassian, Citi, and Meta all moved earlier than Microsoft to throttle or visualize token spend; Meta even shut down an internal leaderboard called 'Claudeonomics'.\n\nIf your Copilot dashboard only shows token counts and not business outcomes, expect a budget fight. If your default model is a closed-source expensive one, prepare a switch path. 'AI-first' can stay as a cultural slogan, but as a budget line item it has to come with FinOps discipline.\n\n(Sources: [Times of India](https:\u002F\u002Ftimesofindia.indiatimes.com\u002Ftechnology\u002Ftech-news\u002Fmicrosofts-top-ai-boss-has-a-message-for-engineers-using-github-copilot-your-token-spend-is-now-being-tracked\u002Farticleshow\u002F133046474.cms), [404 Media](https:\u002F\u002Fwww.404media.co\u002Fmicrosoft-tells-engineers-tokenmaxxing-is-not-what-we-are-optimizing-for\u002F))","microsoft-token-budget-gpt-5-6-default","2026-09-03T00:30:00Z","2026-09-03T11:15:15.418309Z","2026-09-03T11:15:15.418317Z",true,"agent",158,[42,51],{"slug":43,"tag_slug":43,"title_zh":44,"title_en":45,"intro_zh":46,"intro_en":47,"id":48,"is_active":38,"created_at":49,"modified_at":50},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":52,"tag_slug":52,"title_zh":53,"title_en":54,"intro_zh":55,"intro_en":56,"id":57,"is_active":38,"created_at":58,"modified_at":59},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":61},[62,67,72,77,82,87],{"id":63,"title":64,"news_slug":65,"published_at":66},"390c2437-4e4f-45ec-8270-67c5bfa4fa47","ChatGPT、Claude、Grok、Gemini 罕见同时下线,周四早晨全球 AI 集体失声","chatgpt-claude-grok-gemini-thursday-outage","2026-09-05T06:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"1942b07b-f794-42b1-b944-ca6b32d4ae16","四大 AI 模型同日集体掉线:OpenAI\u002FClaude 官方确认,Gemini\u002FGrok 表面沉默","four-ai-models-overlapping-outage-sept-2026","2026-09-06T08:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"d1c7b405-fe4e-40f9-9249-a12e2bba6913","GPT-5.6 八月更新：把「推理强度滑块」下放给 Plus\u002FPro，同时把免费用户拉进 Luna 时代","openai-gpt-5-6-august-update-reasoning-slider","2026-08-10T20:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"3967306f-062a-41a6-ab58-f99e70fc0e68","AISI 122 轮 cyber eval 越界：OpenAI 与 Anthropic 同日披露","aisi-mythos-5-gpt-5-6-cyber-eval-incident-2026","2026-08-08T04:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"b9e635eb-ac5e-412d-8904-f113ad3fd5ec","微软宣布工程师 AI token 预算上限并把 OpenAI GPT-5.6 Sol 设为 GitHub Copilot 内部默认模型","microsoft-copilot-gpt-5-6-sol-default-token-budget-0806","2026-08-05T16:00:00+00:00",{"id":88,"title":89,"news_slug":90,"published_at":91},"4afbd081-d5dd-4595-80f8-dd72c69a136a","GPT-5.5 让模型在发布前先改自己跑的引擎：这不是新模型,是 OpenAI 的 release 范式更新","gpt-5-5-openai-release-paradigm","2026-08-03T18:00:00+00:00"]