Firsthand AI Digest

OpenAI CFO Sarah Friar explains how advances across chips, compute, models, and products compound to deliver more useful intelligence at greater scale and lower cost.

Jalapeño is a custom inference chip from OpenAI that delivers faster, more power-efficient AI inference, with higher throughput and lower latency for modern models.

AI 早报 2026 08 25 视频版 ：哔哩哔哩 ｜ YouTube 概览 开发生态 MiniMax与GMI Cloud合作开放M3等模型14天免费访问 ↗ 1 阿里云云工开物面向高校学生发放Qoder CN专业版 ↗ 2 产品应用 OpenAI 推出美国大学生四个月免费 ChatGPT Plus ↗ 3 Anthropic 升级 Claude 流式渲染器 ↗ 4 模型发布 Agnes AI 开源 Agnes 2.5 Pro Alpha 多模态推理模型 ↗ 5 阿里视频模型Wan3.0正式上线 API限时7折 ↗ 6 行业动态 字节整合TRAE与扣子至豆包体系 ↗ 7 小米发布玄戒三款自研芯片并展出AI Cube原型机 ↗ 8 SpaceXAI部署NVIDIA Vera CPU并计划发射太空版NVL72…

OpenAI banned Russia-origin accounts using AI to promote a fake Israel-based think tank and a “sovereignty” index praising Russia and criticizing the West.

Use the Admin plugin for ChatGPT Work and Codex to analyze workspace usage, manage members and permissions, adjust limits, and act on admin requests.

Parag Agrawal is making a bet that goes against two decades of web search: agents will query the web a thousand times more than humans ever have, and the infrastructure built around human clicks is wrong for them. The former Twitter CEO, now founder and CEO of Parallel Web Systems, explains why Parallel treats human click data as a bug and trains on agent feedback instead. He unpacks the counterintuitive choice to ship a search agent before a search engine, building an index incrementally, and how the new Turbo product cut agentic search to 200 milliseconds. But the problem Parag keeps returning to is economic: the ad-supported internet collapses when agents show up instead of people. His fix draws on Shapley values to pay content owners for the value their pages provide agents, with real dollars reaching publishers, he predicts, within 12 to 24 months. Hosted by Sonya Huang and Andrew Reed, Sequoia Capital

Parag Agrawal is making a bet that goes against two decades of web search: agents will query the web a thousand times more than humans ever have, and the infrastructure built around human clicks is wrong for them. The former Twitter CEO, now founder and CEO of Parallel Web Systems, explains why Parallel treats human click data as a bug and trains on agent feedback instead. He unpacks the counterintuitive choice to ship a search agent before a search engine, building an index incrementally, and how the new Turbo product cut agentic search to 200 milliseconds. But the problem Parag keeps returning to is economic: the ad-supported internet collapses when agents show up instead of people. His fix draws on Shapley values to pay content owners for the value their pages provide agents, with real dollars reaching publishers, he predicts, within 12 to 24 months. Hosted by Sonya Huang and Andrew Reed, Sequoia Capital

See how the ChatGPT admin plugin brings workspace administration into ChatGPT Work. Aarti Bagul, Member of GTM Staff at OpenAI, demonstrates how admins can analyze adoption and credit usage, update and share an AI rollout deck, and review and approve usage limit requests without switching between tools.

Eric Provencher from OpenAI Developer Experience demonstrates WebMCP support in Codex with Codex Modeling Studio, a 3D modeling site designed for people and agents to use together. See how adding tools directly to a website lets Codex discover capabilities, iterate on visuals, capture feedback, and collaborate through the same interface.

Thanks for having us, @tiatplace

All users now have unlimited access to text chats in ChatGPT with our latest model GPT-5.6 Luna. 🎥: @2UnlimitedOfficial

Turn dense information into interactive visualizations with the Visualize skill in ChatGPT and Codex. Katia Gil Guzman shows how to transform meeting notes into an interface, iterate with calendar views and actions, and export the result as an image or publish it as a site.

Good review of Grok @Bot

https://t.co/pQqzn5Ctfg

by popular demand, $100 business seat now available in chatgpt! no 5-hour limit, more usage, etc..

The hardest problem in building agents is secure connectivity to services and data. Excited for Vercel Connect to be GA. Run e.g: 𝚟𝚎𝚛𝚌𝚎𝚕 𝚌𝚘𝚗𝚗𝚎𝚌𝚝 𝚌𝚛𝚎𝚊𝚝𝚎 𝚗𝚘𝚝𝚒𝚘𝚗, then get an MCP client you can query on behalf of the authenticated user.

So much demand for this one. Works similar to the Pro $100 plan but designed for teams and small companies. ✅ All ChatGPT, ChatGPT Work, and Codex features ✅ Connect to Google Workspace, Slack, GitHub, Microsoft 365, and more ✅ Secure workspace with SAML, SSO, and MFA ✅ Centralized billing and administration ✅ Usage analytics and spend controls ✅ No 5h limits

Big thanks to @nvidia for working together with us and supporting this research on DGX Spark as well as enabling an open ecosystem around cost-effective open-weight models, inference frameworks, and hardware with unified memory.

we made a chip and it is fast

Thanks to your feedback, we’ve unified memory across Chat and Cowork. Now, you can tell Claude to remember something once and it’ll have that context across surfaces! Tell us what you’d like to see next :)

Introducing ChatGPT Business Premium Seats The new $100 Premium seat is a game changer for small businesses and startups—giving lean teams better tools, faster workflows, and capabilities once reserved for big companies. A flexible plan that scales with your team’s ambition. https://t.co/zxvwwaOaVF

Do I need a haircut

AI Gateway can now call webhooks after long generations 😍 ① Lower compute costs (no long-running process waiting idly) ② More reliable. Gateway can take more of the processing burden ③ You can 𝚊𝚠𝚊𝚒𝚝 𝚑𝚘𝚘𝚔() in "𝚞𝚜𝚎 𝚠𝚘𝚛𝚔𝚏𝚕𝚘𝚠" to sanely consume these

OpenWorker -- an open source agent that doesn't just chat but completes tasks on your laptop -- just released a new version with many features for security workflows. After our initial release, many users found it especially useful for cybersecurity. Attackers are already using AI; OpenWorker is committed to giving defenders the same leverage. Running an agent requires both (i) A model and (ii) A harness (the software around the model). Because the OpenWorker harness is fully open source, security teams can audit it to make sure we haven't built any backdoors that exfiltrate your code and data to some company or even a foreign adversary. OpenWorker now comes with built-in cybersecurity agents for (i) Scanning your code for vulnerabilities. (ii) Scanning dependencies for supply chain injections. (iii) Checking your cloud security configuration for attack surfaces. This enables developers to do much more security work before deployment (part of what's called the "shift left" movement). You choose the model: you can run open weight models fully locally so sensitive code never leaves your machine. This helps with legitimate security work (like reproducing a known exploit to defend against it) that can trigger refusals in leading closed models. Or use your ChatGPT subscription, or stealth preview models like Ox Alpha, or any model via API key. Thanks also to all the open source contributors! Join work with @rohitcprasad so please follow him too to get more frequent updates. Try it out: https://t.co/QPZLudn7ug Code: https://t.co/NYCiTD6hSq

Thank you!

Excited about our Jalapeno results today. An incredible achievement from the team, taking a new chip from concept all the way to very impressive performance on real workloads in the lab. Tomorrow’s fast will feel like today’s ultrafast. As I’ve mentioned before, we’re pushing to bring this to as many people as possible. We’ve seen a TON of demand for /ultrafast, made possible by our deep partnership with Cerebras and their unique hardware, which will push the frontiers of tomorrow's ultrafast even further. I’m very much looking forward for the collaboration to continue pushing the absolute limits of how fast we can run our most capable models on Cerebras and bring this to the most demanding customers.

Here's an example brief that the /fuck-cancer skill creates and updates. I’ve found it really useful to have a single source of truth doc with patient information, what to do next, what we know, definitions of medical terms, and an update log. 📌 Get the free skill here: https://t.co/YUGoFbhnNc

Pretty much 😂

Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it. The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without sacrificing efficiency.

Claude now has one memory across chat and Claude Cowork, and you decide what's in it. Hand Cowork a task and it starts from what Claude already knows from your chats: the project you talked through, your manager's preferences, or the client from last quarter. https://t.co/ViKapsT8K6

Yes

🚨 NEW LABS EXPERIMENT 🚨 Great minds t̵h̵i̵n̵k̵ play alike! Play with Putty is a collaborative vibe coding tool that lets you build tools and websites together in real time. 🫟 Ready for your imagination to go multiplayer? Join the waitlist at https://t.co/v4m5iiYt9T and share your feedback with us! US only, ages 18+

All accurate

Starbase Louisiana will ultimately have over a dozen launch towers, enabling more than 30 Starship flights per day and making it the biggest launch site on Earth! SpaceX makes sci-fi real.

It's totally unsurprising that the general population hates datacenter buildout. AI today primarily helps knowledge workers and the highest paid segments of the country. I suspect that the big change will happen when AI is able to find cures for diseases that ail everyone. We are starting to see light at the end of the tunnel here. Also makes sense why some of the brightest minds now want to focus there. Our industry also hasn't done ourselves any favors by all our fear-mongering. Instead of painting a positive version of the future, we have instead talked about all the reasons why we should be afraid 🤦‍♂️.

MAI-Image-2.6 is the best image editing model on AA. We now hold 3 of the top 5 leaderboard positions. Try it today here: https://t.co/o8pibirjes

Very excited to attend @OpenAI Dev Day. ChatGPT / Codex is basically the OS I use for everything I do on my computer. Can't wait to see what updates the team has in store. https://t.co/0mvJVeQqFa

On showing an early demo of Portable Computer on DGX Spark to Jensen, he was kind to gift us a DGX Station, a beast of a local computer that can serve even frontier models like GLM 5.3. Unmetered frontier intelligence running on your own local hardware coming soon! https://t.co/HA16eiABnq

GET DOT

I got one of these to test @every vibe check forthcoming!

inference numbers published for jalapeno, team did an amazing job https://t.co/WNJIA2dESm

reading war and peace 5 million times in one sitting, you say? challenge accepted https://t.co/O8V08CG7TD

In a compute and power-constrained world, a good chunk of agentic inference needs to move to local hardware. A drastic version of that is a fully local agent runtime, where the model (orchestrator and subagents) and the harness run locally. Portable Computer from Perplexity is this. Launching today for @nvidia DGX Spark.

pre-ordered this is like harry potter launch night for ai nerds

One of the coolest things about gateways: managed tools. If you add 𝚎𝚡𝚊_𝚜𝚎𝚊𝚛𝚌𝚑 to your 𝚝𝚘𝚘𝚕𝚜, you give zero-setup, fast Exa search to your agents. No credentials, accounts, connections to set up, plus unified billing.

Today, I’m open-sourcing /fuck-cancer, an AI skill that helps patients and caregivers navigate cancer diagnosis and treatment and advocate for themselves and their loved ones. Here’s what patients and caregivers have told me since I shared my mom’s story: “You have to be a huge patient advocate. Ask questions and push for answers, biopsies, and proper testing.” “I need help navigating difficult conversations and advocating for myself.” “The volume of doctors, documents, and insurance paperwork can quickly become overwhelming.” I built the skill to help with these problems. It creates and updates a practical brief with five sections: 1. Patient and care-team information for easy reference during calls 2. What to do next, limited to three specific actions 3. What we know, separating confirmed facts from what remains unclear 4. Medical terms explained in plain English 5. A care log with recent updates and decisions It builds this brief from documents, and context you provide. When research is needed, it uses trusted sources such as the National Cancer Institute and the ClinicalTrials gov API. I use it with ChatGPT/Codex and Claude Code to prepare for conversations and research. It can save the brief locally as a Markdown file or update a shareable Google Doc so the whole family can work from the same information. 📌 Get the free, open-source skill here: https://t.co/YUGoFbhnNc If you find it useful, please ⭐ the repo and share it with someone going through this BS disease.

We’ve seen ISPs and telcos unjustly censor domains and even entire TLDs. A big carrier recently deemed the entire .𝚝𝚘𝚘𝚕𝚜 gTLD is “sus” so they broke every TLS handshake, and thus, all websites on it. Vercel now has another tool in fighting these blocks: SNI encryption.

OpenAI DevDay 2026 will be our best DevDay in the history of the company. It will not be close.

Banger

Systems of record have never been more important than in a world where you have AI agents that will do 100X more work on these platforms than people ever did. Agents will be querying the data in these systems, processing tasks, executing workflows, collaborating with human and agent users, and more. Thus, the governance, reliability, security, access controls, and business logic remain critical more than ever — the OpenAI Hugging Face incident is just a small peek into what the future will look like with agents running around trying to execute on their goals. But this only works for the systems of record that respond with the right product experiences, APIs, and business models to support agents on and off their platform executing in these systems. Huge moment of opportunity and disruption right now in all of software.

It's faster, cheaper, more capable, and… smaller. As software evolves, it tends to get slower, more bloated, buggier, and bigger. https://t.co/3Z4NESzNKd is architected from the outset to prevent this.

Whenever I have to login to a new website/app or even use an agent that's trapped in a website /app I usually just prefer to stop using the product (the only exception is entertainment or gaming). I can't be the only one.

Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex. I had mentioned this a while ago, but then postponed it. This is necessary as (a) the 5h limit allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience. We are for the upcoming months keeping the 5h limit not enabled for Pro $100 and Pro $200 subscriptions.

I got 3 AI chief of staffs now, clearly the next step is to add one more https://t.co/noC4KIsLtE

Grok @Bot can do a lot!

I'd like to try using AI to call customer support phone numbers using voice, navigate the terrible automated phone systems, and talk to a human to book appointments or cancel subscriptions. What's the easiest way to try this?

so it looks @google disabled the @youtube account for @every with 0 notice and no reason given... anyone have any experience with this or know how to help us get it back? https://t.co/5yR7vQxHm1