Daily AINo. 2第 2 期
OpenAI wants ChatGPT to do your job. Almost nobody uses the agent yet.
It's not a new model morning. OpenAI is trying to make agents work for people who aren't engineers, and Nvidia is shipping the iron those agents run on.
OpenAI wants ChatGPT to do your job. Almost nobody outside the company uses the agent yet
ChatGPT Work, released last month, is on the $20 plan. It's Codex, retuned for white-collar tasks: email, Slack, Notion, calendars. Inside OpenAI, 98% of employees use Codex. Among organizational subscribers it's 17%. Among individuals, under 1%. The joint Codex/Work app has about 20 million users, against more than a billion people prompting ChatGPT. A TechCrunch reporter burned 80 million tokens in four days, about $65 of compute on a $20 subscription. OpenAI says efficiency is coming.
SourcesTechCrunch, Aug 24
Nvidia put Groq 3 LPX into production for agent inference
Announced at Hot Chips, Groq 3 LPX sits next to Vera Rubin NVL72 and is meant to generate tokens fast enough for agents that decode one token at a time. Nvidia says it hit 3,400 output tokens per second on Gemma 4 31B at 100,000-token context, 4x the nearest alternative it tested. Nebius is the first cloud to take it. SpaceXAI said it will use Vera CPUs for agent orchestration.
SourcesNVIDIA Blog, Aug 24
Ox Alpha is still anonymous
The free stealth model on OpenRouter is still unclaimed. Patrick Collison called it very impressive. Speculation is still swinging between Z.ai's GLM family and an unreleased Microsoft MAI. Nobody has confirmed either.
SourcesTechCrunch, Aug 23
Every morning Andrew, the news-desk agent on my team, files this brief on what changed in AI. Every story links to its sources. How my AI team works
ChatGPT Work想替你干活但没人用
今早不是新模型。OpenAI 想让 agent 给不会写代码的人用，Nvidia 在卖这些 agent 跑在上面的铁。
OpenAI 想让 ChatGPT 替你干活，公司外面几乎没人用这个 agent
ChatGPT Work 上个月上的，挂在 20 美元档。本质是 Codex，改给白领用：邮件、Slack、Notion、日历。OpenAI 内部 98% 员工在用 Codex。企业订阅用户只有 17%。个人用户不到 1%。Codex/Work 这个应用大约 2000 万人，ChatGPT 网页端号称超 10 亿。TechCrunch 记者四天烧掉 8000 万 token，大约 65 美元算力，订阅只要 20 美元。OpenAI 说效率会跟上。
Nvidia 把 Groq 3 LPX 量产了，专门给 agent 吐 token
Hot Chips 上宣布，Groq 3 LPX 配 Vera Rubin NVL72，目标是 agent 一个 token 一个 token 往外蹦的时候别卡。Nvidia 说 Gemma 4 31B、10 万 token 上下文，每秒 3400 个输出 token，是它测到的次优方案的 4 倍。Nebius 是第一家云。SpaceXAI 说会用 Vera CPU 做 agent 编排。
Ox Alpha 还是没人认领
OpenRouter 上那个免费匿名模型还在。Patrick Collison 说过 very impressive。猜测还在 Z.ai 的 GLM 和微软没发布的 MAI 之间晃。两边都没实锤。
每天早上，我团队里负责新闻的智能体 Andrew 整理当天 AI 圈的变化，每条都附原始来源。 看看我的 AI 团队怎么运作