AI 服务重置雷达

my dot stopped showing up in my codex desktop app yeah openai is cooking...

This is the reason why OpenAI won’t win the personal agents war Muse at least realizes that normies want one agent

Why is Codex giving reset every day? 😭

@petergyang @amorriscode hold my beer

With plugin extensions, I can customise my Codex desktop app. Google cal feels a bit too cluttered for me, so I built a simpler focused month and day view, exactly how I like it. Software for n=1

7,000 followers! don’t ask me anything, I work at anthropic

@thsottiaux @ajambrosino I quantized him on Friday

@thsottiaux green

@thsottiaux 👀

Tibo@thsottiaux

Day 13/ Le limit reset

is chat the final form of agent ui? please share why

@thsottiaux 🤫 we'll let the work do the talking

Tibo已经没招了，我来出几招 1. plus取消5h限制 2. dot给plus用户使用 3. ultrafast改为提升15倍速度 4. Pro20x的用量翻倍 5. 提升6.1 Sol的TPS至少到50 这已经够Tibo 5天的了🤓

how do you trust your agents' work more? copy this post/link and give it to your @bot or agent! 1. codebase improvements: this is just good software engineering - better architecture, stronger types and invariants, clear boundaries. agents love to copy what exists and take the shortest path, so make the easy thing the right thing. categorically eliminate problems through code and you won't need to think about it again. 2. static analysis: lint rules, tests, CI, verification (eg TLA+). very token efficient ways to catch issues ahead of time. 3. prose: comments, rules, AGENTS.md, skills - anything that can't be enforced and requires agent judgment. use it minimally, but do use /create-verification-skill. 4. human review: the hardest to scale. at the rate agents land code, it's pretty much impossible to thoroughly review every PR, so spend that effort on the highest leverage point, designing an environment where agents do the right thing by default. this forms an inverse pyramid: you want most of your effort at the top, least at the bottom. codebase improvements are deterministic, making it more token efficient, and lets you sleep more soundly at night. install https://t.co/u9xIlN3eN0 and create a new Project: 1. look through my slack/chat messages and code review comments where i've given feedback on PRs, code, architecture 2. use /recall on my past transcripts to learn how i steer agents 3. /poteto-mode use /correct and /architect on past commits to find the most common pitfalls agents fall into, then propose a plan to rearchitect the code i own so it's more agent friendly. answer open questions with prototyping rather than defer to me, and back the plan with real data. 4. before you begin, use /sharpen + /in-your-own-words and confirm this prompt with me

Dots are for doing. Chat is for learning and discussing. Work is what you use if you have no Dot.

See, @OpenAI? It’s fine to build hype. But then you gotta deliver. Spot the difference.

🗣️ THIS WEEK IN AGENTS API 🗣️ ⚡ 6-Astra Ultrafast: now available in the Agents API! 📷 Sandbox snapshots: Restore OpenAI-hosted environments after idle with files intact and keep the work going. 🤑 Better billing: you won't pay for environments that are idle 🔑 Environment keys: Create environment keys programmatically (not just through the dashboard) 📈 Artifact SDK: Create and edit spreadsheets, slides, and Word docs in OpenAI-hosted sandboxes 🔎 Turn-level history: Fetch the items from a specific root-agent turn, including subagent interactions 🔒 Vault filtering: Filter vaults by metadata

Tibo@thsottiaux

What is ChatGPT Space, that is the question. Endearingly awkward at times because it's the actual team working on it. What would you like to see next?

Day 12: Dots on the Home Screen Clear improvement, or meh? 👏 or 🤷 here: https://t.co/YiyilRWK8W https://t.co/1VyD7vrNkr

Most common format of DMs I get: “Hey Theo, I’ve been a fan for a long time [3-8 paragraph life story] So, which $20 plan is better, Claude or Codex?”

OpenAI, I think you need to explain this a lot better. The difference is you can use chat mode not just work mode. Why are AI labs so bad at explaining their own tools and systems? 🤷‍♂️

Tibo@thsottiaux

Day 12/ We’ve made improvements to Codex & Work that increase the effective usage you get. We achieved this by reusing context across subagents, keeping context cached longer, and improved request routing that allows us to increase cache hits.

Tibo@thsottiaux

Day 12/ We’ve made improvements to Codex & Work that increase the effective usage you get. We achieved this by reusing context across subagents, keeping context cached longer, and improved request routing that allows us to increase cache hits.

i'll die on this hill-- gpt-live is the most natural voice AI for max naturalness w prompt-to-voice: 1/ specify accent, timbre, and gender in prompt-to-voice 2/ generate 4+ voices before picking your fav 2/ prompt for lots of backchanneling 4/ leverage custom voices

you can just build stuff now i built a digital version of my house in blender based on polycam lidar scans, then astra did a detail pass based on photos. basically every lamp, door, plant or electronic component is viewable in this format. i can toggle lights, see if plants need watering. The blender version is used by imagegen to create these stylized dioramas, that react to their actual state and time of day. this is all seamless inside of the home assistant app as a dashboard. when people talk about personalized software this is what I envision.

you can just build stuff now i built a digital version of my house in blender based on polycam lidar scans, then astra did a detail pass based on photos. basically every lamp, door, plant or electronic component is viewable in this format. i can toggle lights, see if plants need watering. The blender version is used by imagegen to create these stylized dioramas, that react to their actual state and time of day. this is all seamless inside of the home assistant app as a dashboard. when people talk about personalized software this is what I envision.

I never use my iPhone’s camera button. So I made it call my dot. We just shipped dot widgets and Action button support on iOS. Here’s how to set those up, plus the Shortcuts trick I used for the camera button.

Here at the @OpenAI DevDay Exchange in BLR 🔥🔥🔥

OpenAI Developers@OpenAIDevs

Put your dot on call. Ali, an engineer at OpenAI, used his dot to triage an on-call incident at 3am. His dot investigated the issue and started a Codex thread, so he could steer the work from his phone in the ChatGPT app.

OpenAI Developers@OpenAIDevs

Here’s how our engineers use Codex at OpenAI to dig through our codebase, find what needs attention, and chip away at the maintenance backlog:

Software maintenance is changing. A few months ago, I had to tell Codex exactly where to look for dead code. Now it can find candidates in Slack, check them against production telemetry, and draft cleanup PRs. Here’s how we do this at OpenAI. https://t.co/PdQl43nLeQ

Browser and computer use got waaaay easier in the latest @OpenAIDevs Python and Node.js SDKs 🧵

Great people, great conversations, and exciting possibilities! 🤝 Connected with @KushalVijay_ at #OpenAlDevDayExchange , Bangalore. Here’s to building, learning, and innovating with AI! 🔥 @OpenAIDevs @OpenAI #AI #Technology

Had an absolute blast at the @OpenAI DevDay Exchange in Bengaluru! 🔥 Met so many new people and friends, had great conversations with the OpenAI team, and the demos, booths, and talks were all next-level. Learned a ton!!! Seeing the old ChatGPT UI from the early GPT days during @derrickcchoi session hit different though We’ve come such a long way from those simple conversations to the powerful agentic workflows and tools we have today. Wild how far this has come. 🚀 Grateful for the energy and the people.

Submitted the game I built on ModRetro Chromatic with Codex. 🫶 https://t.co/7DWtnR8jM6 Now I cannot wait to build more hardware and actually but my hardware engg degree into use🤓

Open AI Devday '26 Designers attending an engg conference is a normal thing now. Lovely talks all around and learned a few new codex tricks.. Shout out to @mvkaran for the Free Codex Pro as well ;) Brb gtg tokenmaxx...

Had a blast at the @OpenAI Devday !!

cool people at @OpenAI dev day @the_dream_saver @tanishasharmax @joeyousss @vedantstwt

OpenAI Developers@OpenAIDevs

Building apps with GPT-Live-1? Give them a voice that fits. Here’s how to shape a voice with a prompt or choose from new international preset voices:

.@OpenAIDevs DevDay Exchange Bengaluru is here!! 🎉 Say hi if you are around

That’s a wrap on OpenAI DevDay Exchange Bengaluru 🇮🇳 Loved meeting everyone and hope to be back soon!

Codex Ambassadors less go!! Pleasure to grow this community in India 📍OpenAI DevDay, Bengaluru @OpenAI @OpenAIDevs @reach_vb @gabrielchua @paw_lean @sama

I really need to get more creative Was such a pleasure to meet so many cool folks at DevDay Exchange Bengaluru

That’s a wrap on @OpenAI devday 2026 bangalore! @OpenAIDevs

interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works https://t.co/zHij5zwKGT

@OpenAIDevs And can I upload my beta versions there?

@danizeres Love this! https://t.co/5TXNDPj7gH

You can just build [physical] things

I miss the old Tibo, reset the usage Tibo The “oops, again” Tibo, back to a hundred Tibo I hate the new Tibo, arguing in quotes Tibo The 28 days Tibo, beefing with poteto Tibo I miss the sweet Tibo, free weekend reset Tibo I gotta say, back then I’d never hit my weekly, Tibo See I believed in Tibo, there weren’t any Tibos And now I look around and there's so many Tibos

The Codex name has been incredibly sticky (all of the technical folks I know still call the desktop app Codex), but clearly ChatGPT is the long-term product Increasingly feels like Codex should be the brand of the harness: ChatGPT powered by Codex Agents API powered by Codex Dots powered by Codex

https://t.co/tZZoXESVed

My dot sent 60 messages to movers over several days, filled out quote forms, and handled the back-and-forth. I gave it my dates, what I needed moved, and permission to handle the conversations. Wrote up how it went, and how to give yours a job like this. https://t.co/UsAlaLh7gj

I'm thoroughly Dots-pilled. Still a fan of Muse, but probably 75-80% of my AI requests now go through Dots. The primary exception: learning things in chat. For that, GPT-6 Instant with Dynamic UI is unmatched. Just incredible. Some recent Dots work and personal tasks: - Build and update a Spanish vocabulary app - Maintain a document on AI comparisons - Execute tasks I delegate directly from Slack - Monitor threads in Slack and summarize responses - Research a contact and write a brief for a call - Analyze token data and make me a deck to share - Monitor token consumption, alert me if needed - Respond to comments in Pages in Space You might say: "Dude, ChatGPT could do a lot of that already." True. But I no longer have to think about all these threads individually. I just keep firing stuff at my Dots from wherever I am. ChatGPT, Slack, reading a Page in Space. I just throw stuff at it, and it does it, and sometimes it proactively just does stuff based on what it thinks I'd find valuable. I was skeptical when this launched because it's nowhere near as consumer-friendly as Muse. But I think that's part of why OpenAI launched this first to business and pro-tier users. We're willing to suffer a bit for extra capability. I'm starting to see some of the advantages now, and the product also continues to get more polished. Bullish.

Tibo@thsottiaux

Bonjour

OpenAI@OpenAI

New voices. Better accents. And less "checking..." GPT-Live-1.1 brings these updates to Voice in ChatGPT, Codex, and dots, with new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese. Rolling out today to all users. Coming soon to the API.

OpenAI Developers@OpenAIDevs

A robot that turns photos into drawn portraits: https://t.co/Pjw8PQTQpK

OpenAI Developers@OpenAIDevs

An e-ink live train tracker display: https://t.co/PN3mmZTe2L

OpenAI Developers@OpenAIDevs

Pitch your dreams and save them with this dream logger: https://t.co/tdHYV59pqT

OpenAI Developers@OpenAIDevs

You can just build things with Codex We built a site for you to explore hardware projects from the community and share your own.

We’re also removing the cap on how many Sites you can create on ChatGPT Plus and Pro. We've been listening and we want to keep making it easier for you to build whatever you want. Have fun building ChatGPT Sites!

OpenAI@OpenAI

We're expanding access to GPT-Rosalind for academic research. The new Rosalind Academic Plan helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next—with sponsored access for eligible labs at U.S. universities. Academic research starts with questions whose impact may take years to impact human health.

OpenAI@OpenAI

GPT-Rosalind is built to support that work: helping researchers reason across scientific literature and biological data, with scientists bringing the expertise and judgment to evaluate the evidence. Eligible labs can receive up to 12 months of sponsored access to OpenAI tools, with GPT-Rosalind access subject to trusted-access approval. Research data shared through the Rosalind Academic Plan is not used to train our models by default.

OpenAI Developers@OpenAIDevs

Use workspace custom domains to publish on your organization’s verified domain, like “support.sites.[yourdomain].com” Private tunnels connect hosted Sites to approved internal infrastructure and APIs.

OpenAI Developers@OpenAIDevs

Sites is now generally available for Enterprise, with US and EU data residency for new Sites in eligible workspaces. Build team apps with databases and authentication. @kagigz shows how to schedule updates and let teammates use their own connected data with Plugins in Sites:

ChatGPT@ChatGPT

Try using ChatGPT in Chrome for tasks like… “Compare the three couches I have open by dimensions, material, delivery time, and return policy.” “Explain this article in plain English. What’s the main argument, and what evidence supports it?” “Compare these pricing pages against our requirements. Which plan includes everything we need?” “Use the details in this event brief to fill out the registration form, and stop before submitting.” “Put the vendors in these tabs into my comparison sheet, with pricing, key features, and source links.” “Look at all the tabs I have open and compare the contractor services. Then draft an email to the one that fits my home renovation.” “Fill out the form form I have open in @chrome ahead of seeing my new dentist. Also, can you add the appointment to my work and personal calendar.”

ChatGPT@ChatGPT

Next up: using ChatGPT right within a Chrome tab got a major upgrade – the full power of ChatGPT is now in your browser. Your entire ChatGPT history is right there, and you can ask question in Chat or get things done in Work – side by side while you’re on a site.

I'd make an excellent CCP honeypot for Ant/OpenAI researchers, if it weren't for my obvious shortcomings 😥

Me: “show me your @chatgpt account” 12 yr old Son: “are you sure you want to read my chats?” /proud-father

@aileenvl @OpenAIDevs @paw_lean @rhiannon_io @fedesarquis dammmn me dieron ganas de soldar cosas jaja

@thsottiaux upgrades to the auto-complete or an orchestrator massively slept on. I actually think that this was probably the best thing you've shipped I don't think you guys properly explain what that does. It's basically replaced my... What do I do next?

ChatGPT Sites + MCP Events + dots is a really fun combination. Here’s a toy example: I submit an image request through a Site → the Site sends an MCP event → my dot picks it up and generates the image. Imagine the same flow for bug reports, support requests, or new sales leads. Build a Site around your workflow, send updates to your dot, and tell it what to do when they arrive. @OpenAIDevs

How can the math community forge the best path forward in the AI age? We're committing $15M to workshops and initiatives — plus free AI computing resources for foundation-supported researchers — through new partnerships with @AnthropicAI and @OpenAI: https://t.co/G8slrHoiad #math #science #AI

@thsottiaux - Skills that route to different behaviors based on model - Skills that have recs like "don't use slow git command" when using ultrafast mode - Attribution in things like github prs (this only works because I shim it in on T3 Code side)

Tibo@thsottiaux

Landing within the next 2 hours

It is so stupid that Codex STILL does not include the current model in the system prompt. Your agent has no idea what it is beyond "based on GPT-6". I've complained about this for a year straight and they won't fix it for no good reason.

I had @bot build a tier list of how positive my interactions with AI creators are, fun experiment. Not sure why it put @theo in his own tier 💁

@thsottiaux @poteto @theo can't tell if you're running or they're your army

Tibo@thsottiaux

The reports of my reset button being lost are greatly exaggerated Day 10/ Reset

@theo @thsottiaux @thsottiaux come back

Anyone check in on @thsottiaux recently? Over 24 hours without posting is concerning. Hope he's okay

🚀 Introducing Next-Concept Modeling (NCM): our new paradigm for fast, parallel, coarse-to-fine text generation. Compared to strong diffusion language models (DLMs) at similar parameter size, NCM achieves: ⚡ Up to 31× faster generation on downstream tasks like summarization and machine translation with comparable or better quality. 🏎️ 10–21× faster unconditional generation across TinyStories, LM1B, and OpenWebText, with comparable or better generative perplexity and using 3–5× fewer training tokens. 🧩 Better performance on reasoning tasks like Sudoku, outperforming both AR and diffusion baselines with fewer sampling steps. With @ishanskhare, @IddahMlauzi, @ekellbuch, @simran_s_arora, @Azaliamirh, and @HazyResearch. More details in the thread 🧵👇

Codex ModRetro Chromatic in Vancouver!

I built a dream logger. For all those dreams I forget five minutes after waking up. This little @M5Stack CoreS3-SE turns dreams into images using the @OpenAIDevs image-gen API. 🧵

With plugin extensions, I can customise my Codex desktop app. Google cal feels a bit too cluttered for me, so I built a simpler focused month and day view, exactly how I like it. Software for n=1

Excited to announce Next-Concept Modeling (NCM): our new paradigm for fast and parallel, coarse-to-fine text generation! 🚀 Compared to strong diffusion language models (DLMs) at similar parameter sizes, NCM showcases: ⚡ 31× faster generation on downstream summarization and translation tasks. 🏎️ 10–21× faster unconditional generation across TinyStories, LM1B, and OpenWebText, with comparable or better generative perplexity using 3–5× fewer training tokens. 🧩 Better reasoning performance on tasks like Sudoku. Check out the paper and codebase 👇 Arxiv: https://t.co/NUCOYycgcM AlphaXiv: https://t.co/4brvP8GxDg Code: https://t.co/CnkJHS4KQt Enjoyed working with @KumbongHermann and advised by @simran_s_arora @Azaliamirh and @HazyResearch.

The MOST UNDERRATED Codex update is the "autocomplete" prompt suggestions ... this thing is reading my mind, and adds things I would be too lazy to type and usually reserve for scenarios when the concise version might do. My prompts are better than ever now ... cudos where it's due @thsottiaux - congrats to everyone involved!

I couldn't make @OpenAI DevDay this year so I gave my ticket to my colleague @patrickerichsen as he's a SF local. In return, he promised me the swag he acquired, if he knew what kinda swag it was gonna be, he might of not made that promise 🤣 @modretro

OpenAI Decisions API has 30x the Jev's context. But you don't have to use it all. An 8k context budget was enough to classify financial documents at 98.25% F1—with 84% lower model cost than our 480k reference.

GBC is a bit hard to work on. It’s lacking the colors I want since it can only handle so much. I asked Codex ModRetro Chromatic plugin to port over the art and some gameplay to GBA instead and 🤩 Here’s a side-by-side comparison.

ChatGPT just got MUCH better. With GPT-6.1 Sol as the default + recent memory upgrades, voice chats have sharper humor, emotional nuance and persistent older memories. Best used w/Standard Voice on iPhone or Android. Please keep it, @OpenAI 🔥 🔥 🔥

📷 Custom(AI) GoPro Ingestion Tool 🤖 Trying to streamline the import process for the 4 GoPro array that I am working with. Asked #Codex to help me find both the quickest as well as least friction ways to ingest footage. So I'm now able to plug them in to review and import or gracefully have them trickle ingest over wifi when plugged in and connected to the local WiFi. Additionally I am able to inspect the current firmware etc. @OpenAI #AI #Prototype #Productivitiy #Tools #Astra @GoPro #Workflow

📷 Custom(AI) GoPro Ingestion Tool 🤖 Trying to streamline the import process for the 4 GoPro array that I am working with. Asked #Codex to help me find both the quickest as well as least friction ways to ingest footage. So I'm now able to plug them in to review and import or gracefully have them trickle ingest over wifi when plugged in and connected to the local WiFi. Additionally I am able to inspect the current firmware etc. @OpenAI #AI #Prototype #Productivitiy #Tools #Astra @GoPro #Workflow

As AI continues to advance, I think it’s even more important for humans to come together, connect, & build a shared vision for the future. Six weeks into my role on OpenAI's DX community team, I’ve seen that happen through our Codex Ambassador events... https://t.co/7S0AFc4lB1

They're distilling me at work tomorrow

Do you use ChatGPT Sites? What do you build with it? How can we make it better?

can someone check on @thsottiaux i miss him

OMG. I asked GPT-6 Astra to turn @thsottiaux's profile picture into a fully 3D model. I don't think he's ready to see this. 💀 prompt is in the comments

the new codex prompt suggestions are humbling it suggested the exact multi line sentence i was going to say

I gotta hand it off to OpenAI. Their composer predictions is so forward thinking. It's really came up with blind spot suggestions that I might not have thought about or at the very least sometimes a different way to think about my question. Big fan.

all active Stanford students, faculty, postdocs can use chatgpt edu for free here: https://t.co/paaEf8wjm4 no reason not to take advantage

Codex composer predictions really know what I want

For the second time in Google Chrome’s history, Google awarded a $250,000 bounty for a full-chain exploit… to @XBOW. Here’s the full story: At an all-hands at the start of 2026 in Malta, I set the team a crazy challenge. A full-chain exploit of Chrome by the annual Black Hat cybersecurity conference in August. I believed there were few better tests of whether @XBOW could move beyond web applications to desktop software, servers and devices. Chrome is one of the most hardened pieces of software on the planet, used by more than half the world’s population. A full-chain exploit lets an attacker take total control of a machine, where they can read every email, password and bank login the user types. You don’t need to download anything or click a phishing link. Visiting the wrong site is enough. So @moyix, Andres Luksenberg, Baul Lee, and Willem Pinckaers took on the challenge along with a fifth teammate: OpenAI's GPT-5.6 Sol. They built a chain they thought would win, but Google told us they already knew one of the bugs. We were already in Las Vegas for Black Hat and DEF CON, and the team didn’t want to go home without the win. So Brendan and the agents built a new exploit in just 6 days. We submitted it to Google and waited. The news came through that we'd done what should’ve been impossible. A full-chain bounty of $250,000, the first awarded under Google’s 2026 rules. It took just 6 days and $40,000 of compute to construct an exploit that affects a billion users of Chrome on Windows. An intelligence-agency veteran told me his team used to spend 6 months on this kind of vulnerability. My takeaway: the cost of finding truly critical vulnerabilities is collapsing, for attackers and defenders alike. Restricting powerful models is not the answer; GPT-5.6 Sol is already public. But defenders have one advantage attackers never will: they can test their own code first. Assume your software is exploitable. Whoever is hunting your bugs has the same building blocks we did. This is cause for urgency but not panic. XBOW is now developing an autonomous product for testing compiled software (desktop applications, servers, embedded devices), and partnering with the world’s top organizations to design that product. If you’d like to be one of them, get in touch. P.S. Restart Chrome whenever the Update button appears!

For the second time in Google Chrome’s history, Google awarded a $250,000 bounty for a full-chain exploit… to @XBOW. Here’s the full story: At an all-hands at the start of 2026 in Malta, I set the team a crazy challenge. A full-chain exploit of Chrome by the annual Black Hat cybersecurity conference in August. I believed there were few better tests of whether @XBOW could move beyond web applications to desktop software, servers and devices. Chrome is one of the most hardened pieces of software on the planet, used by more than half the world’s population. A full-chain exploit lets an attacker take total control of a machine, where they can read every email, password and bank login the user types. You don’t need to download anything or click a phishing link. Visiting the wrong site is enough. So @moyix, Andres Luksenberg, Baul Lee, and Willem Pinckaers took on the challenge along with a fifth teammate: OpenAI's GPT-5.6 Sol. They built a chain they thought would win, but Google told us they already knew one of the bugs. We were already in Las Vegas for Black Hat and DEF CON, and the team didn’t want to go home without the win. So Brendan and the agents built a new exploit in just 6 days. We submitted it to Google and waited. The news came through that we'd done what should’ve been impossible. A full-chain bounty of $250,000, the first awarded under Google’s 2026 rules. It took just 6 days and $40,000 of compute to construct an exploit that affects a billion users of Chrome on Windows. An intelligence-agency veteran told me his team used to spend 6 months on this kind of vulnerability. My takeaway: the cost of finding truly critical vulnerabilities is collapsing, for attackers and defenders alike. Restricting powerful models is not the answer; GPT-5.6 Sol is already public. But defenders have one advantage attackers never will: they can test their own code first. Assume your software is exploitable. Whoever is hunting your bugs has the same building blocks we did. This is cause for urgency but not panic. XBOW is now developing an autonomous product for testing compiled software (desktop applications, servers, embedded devices), and partnering with the world’s top organizations to design that product. If you’d like to be one of them, get in touch. P.S. Restart Chrome whenever the Update button appears!

For the second time in Google Chrome’s history, Google awarded a $250,000 bounty for a full-chain exploit… to @XBOW. Here’s the full story: At an all-hands at the start of 2026 in Malta, I set the team a crazy challenge. A full-chain exploit of Chrome by the annual Black Hat cybersecurity conference in August. I believed there were few better tests of whether @XBOW could move beyond web applications to desktop software, servers and devices. Chrome is one of the most hardened pieces of software on the planet, used by more than half the world’s population. A full-chain exploit lets an attacker take total control of a machine, where they can read every email, password and bank login the user types. You don’t need to download anything or click a phishing link. Visiting the wrong site is enough. So @moyix, Andres Luksenberg, Baul Lee, and Willem Pinckaers took on the challenge along with a fifth teammate: OpenAI's GPT-5.6 Sol. They built a chain they thought would win, but Google told us they already knew one of the bugs. We were already in Las Vegas for Black Hat and DEF CON, and the team didn’t want to go home without the win. So Brendan and the agents built a new exploit in just 6 days. We submitted it to Google and waited. The news came through that we'd done what should’ve been impossible. A full-chain bounty of $250,000, the first awarded under Google’s 2026 rules. It took just 6 days and $40,000 of compute to construct an exploit that affects a billion users of Chrome on Windows. An intelligence-agency veteran told me his team used to spend 6 months on this kind of vulnerability. My takeaway: the cost of finding truly critical vulnerabilities is collapsing, for attackers and defenders alike. Restricting powerful models is not the answer; GPT-5.6 Sol is already public. But defenders have one advantage attackers never will: they can test their own code first. Assume your software is exploitable. Whoever is hunting your bugs has the same building blocks we did. This is cause for urgency but not panic. XBOW is now developing an autonomous product for testing compiled software (desktop applications, servers, embedded devices), and partnering with the world’s top organizations to design that product. If you’d like to be one of them, get in touch. P.S. Restart Chrome whenever the Update button appears!

This is the first—and only—sketch I gave Codex to build the early prototypes of my tool organizer app. I’m thinking of calling it Tooly. A little spoiler 🙂 It still surprises me that something I sketched over a weekend has become a prototype people are enjoying, suggesting features for, and asking me to turn into a real app. That’s a pretty lovely feeling.

Unboxing a package from @OpenAI 📦 This will go straight into the most affordable version of MOSS the open source litter-picking robot. Watch the video: there are some goodies! Thanks @OpenAIDevs, I am excited to build with Codex!

Tibo@thsottiaux

What would you like with your side of reset? It's done.

Tibo@thsottiaux

Global reset landing by EOD PST time. Enjoy

Gemini

Okay I’m finally gonna go film the Muse Wonder video. What questions do you guys have?

We have two additions to PersonalAgentBench: @hark_labs and @OpenAI's Dots! Dots has taken the lead, with Hark close behind while Gemini Spark moves to third, followed by Instinct, Grok Bot, and Muse. We also doubled the number of tasks this week, adding research, image understanding, editing, and everyday communication. The wider set changed the picture. Gemini Spark was strongest on research but weaker on image and editing tasks. Hark and Gemini tied on tasks involving user authority. Dots was the most consistent across task types, although it was weaker when it had to infer personal context. No assistant led every category; the overall ranking rewarded consistent performance across the full set. This is still an early snapshot, and we’re excited to see how results evolve as we add more agents, expand the tasks, and re-evaluate after product updates. If you’re using any of these six personal agents, follow along by visiting the link in the original post.

if tibo doesn’t ship any big update for codex in this 28 days then it’s the end for openai

2 days ago btw

CodeX 砍了我们一半用量后，Tibo 发起了一个投票，问我们想要 2x 用量回来，还是想要 5h 限制去掉。 结果很多人选了 5h 限制去掉。这下轮到 Tibo 瘫坐震惊了，因为他原以为是人都会选 2x 用量。他们也本打算这样做，投票只是起个造势的作用 🤣 说实话，我也选了 5h 限制去掉。因为限制是实打实在那的，而 2x 用量就算回来了 谁知道你会不会暗戳戳地又给降了呢 😠

Day 8 of the OpenAI sprint: I feel like OpenAI just doesn’t give a damn about its users. For the third day in a row, there’s been no news before 10 PM UTC, when most of the world is already on Tuesday. I think the whole sprint idea has failed, and they’re just afraid to admit it.

Tibo@thsottiaux

Day 8/ Reset

6.1 Sol ultrafast >>> 6 Astra ~ same cost but whole different game with 250 tpm

New research on pretraining data mixing in biology. At Radical we're huge supporters of driving frontier research in AI and sharing it with the community. So much cool work at the intersection of biology, this is the next frontier of AI 🧬🔥📈

1/5 What does it take to scale pretraining to 20+ data modalities at once? Improved data mixing methods are one important ingredient. Today, we share how we scale data mixing to thousands of sources for DNA and natural language pretraining of Omnii, our multimodal genome model.

A small peace offering: I put chief @poteto in Codex 🥔

Multi-label document classification is a great use case for Decision models. OpenAI's Decisions API achieves 98.1% F1 versus the agent’s 93.5% at 1/7 of the cost and 28× lower p90 latency.

I think I found what I needed..... Time to test this out with Codex and see what I can make it do with GPT Live 🖤 I'm going to see if I have a smaller speaker. The speaker native to this unit is big, but this board is perfect for me. It's also only $13 so it's a huge win.

I think I found what I needed..... Time to test this out with Codex and see what I can make it do with GPT Live 🖤 I'm going to see if I have a smaller speaker. The speaker native to this unit is big, but this board is perfect for me. It's also only $13 so it's a huge win.

who in SF is building fun hardware with OpenAI APIs and Codex? think fidget toys, little assistants, weird gadgets i’d love to capture your use cases and chat about what you’re making

Here’s the result of my second run with Codex. The first run turned a reference image into an interactive 3D container-terminal simulation. This time, I asked Codex to audit the system in depth, suggest improvements, and implement them. Harbor now includes: - An experiment lab for saving runs and comparing scenarios against a baseline - Repeatable experiments with matched random seeds across scenarios - Timeline replay and branching, so you can rewind, change settings, and explore a different outcome - Individual container tracking, with a journey history from ship to yard - Yard-crane capacity, queues, and shared-road reservations - Diagnostics showing why cranes and vehicles are waiting You can test whether adding vehicles actually improves throughput, see where queues build up, and compare the results of different operating settings. You can also return to an earlier point in a run, change your decisions, and save the new result for comparison. The second pass also improved the rendering, controls, and layout, with clearer cargo stages and completion milestones.

had a spare ESP32 device so I had Codex create a custom firmware for it I now have this little guy that checks on @t3dotcodes in real-time via its orchestration API it's called T3 Pebble

I used GPT-6 Astra to create a 3D website concept for a wilderness hotel via Blender MCP Agents: Codex + Voyager .so Meshes: 13,698 / Triangles: 10,7M / Scene: 489MB 100% 3D scene

OpenAI Developers@OpenAIDevs

Annotations paired with WebMCP give your agents more context to act on. Check out how @kagigz uses custom annotation controls and suggested prompts to help users edit scenes and update expenses in ChatGPT’s browser.

this is Felix, the detective in a mini mystery game i built for nights when i host dinner. and yes: he lives on an M5StickS3 from @M5Stack. i can ask him Qs about the case, pitch my lil theory, and get clues printed on an actual receipt 🧵 I built it with Codex, ImageGen, Decisions API & GPT-Transcribe:

OpenAI Developers@OpenAIDevs

Already showing people what they can build with Codex? Applications to become a Codex Ambassador are open again. We support you in hosting meetups or sharing what you’ve learned: https://t.co/qNrrcEL7sN

Big update in the OpenAI Agents API: dramatically reduce latency with sandbox pre-warming 🔥⏱️ How it works 🧵

New research on pretraining data mixing in biology. At Radical we're huge supporters of driving frontier research in AI and sharing it with the community. So much cool work at the intersection of biology, this is the next frontier of AI 🧬🔥📈

I’m giving my ceiling AC remote control, and a pair of eyes. With the Raspberry Pi 5 that just arrived, and Codex. The next 4–5 weeks are about making it work. 🔧 My Codex Physical Builds kit is here, so today is Day 1. Inside: a Raspberry Pi 5, a black case with the Codex logo, a cooling fan and CPU heatsink, plus a Lexar 128GB microSD. The board has 40-pin GPIO, four USB ports, and two camera/display connectors to work with. The goal is to control my existing ceiling-mounted fan coil unit from my iPhone, even when I’m away from home. Here’s the plan: - My iPhone sends a command to the Pi. - The Pi sends an infrared signal to the AC. - A camera checks the louvers and status LEDs for a response. Sending the command is one part. Checking what actually happened is the next. Right now, I’m feeding Codex detailed photos of the Raspberry Pi, AC unit, and remote to work out what this hardware actually needs: compatible IR transmitters and receivers, a camera, and the cables and components to connect everything. Once the parts are ready, I’ll work with Codex on the control code, test the signals on the actual unit, and see how reliably the camera can read its state. One small computer, an existing AC, and 4–5 weeks to connect the pieces. Time to build. @OpenAI @OpenAIDevs

Low latency decisions with visual inputs with the new @OpenAIDevs Decisions API 🔥

heading to Tokyo for the first time, so my dot and i are building out my first Tokyo trip in Spaces we got maps for each activity, with food (duh!) + coffee stops nearby. planning to work early mornings + evenings so i can spend my days exploring Tokyo and meeting devs while i'm in Asia

I built GTA Toronto + Uber driving with Codex for my new Modretro Chromatic that i got from OpenAI devday two weeks ago. the most fun ive had in year. pure GTA II vibes. game is open source here: https://t.co/k5RCOd61Ob

Started the day rebuilding the tracks. The spacers were a bit confusing, so I asked Codex for a track assembly tutorial with a symbol system. That’s the kind of thing Codex can make for me even when my day is already full. It’s a good feeling to get this extra help, feels a bit like having a superpower. The future really delivers guys 😀 Next: now that MOSS moves, I’ll stress test the mechanics and the electronics, and bring in more dimOS modules for the next logical step, MOSS’s @dimensionalos integration. Today it ran just a few modules, for gamepad teleop.

Lately I’ve been learning math with ChatGPT (Astra) as my tutor, using this document scanning camera and a vibecoded tool that pastes the captured image right into the chat. This way, I just work problems on paper and ChatGPT gives feedback. It’s really nice.

@thsottiaux Now I’m confused. Yesterday you said the opposite of what you’re doing today.

WHAT ARE YOU ALL EXCITED ABOUT WE LITERALLY JUST GOT SCAMMED WTF It has always been the case that we can edit the context window. And now instead of a reset we get "default context lenght went up by 80%"? It was always editable bro...

Codex is way better (now) than Claude code. Who else agrees?

Tibo@thsottiaux

Pro users, would you rather have

just ratiod the reset man (tibo) are you proud of me @poteto

Tibo@thsottiaux

Day 7/ We have increased default context length for GPT-6.1 Sol to 500k when used through Codex & Work, up from 272k or an ~ 80% increase. We will go higher in the future, but it helps reduce occurrence of compactions and with tasks that require keeping a lot of context and nuance over periods of time. It was carefully tested and is the new sweet spot for most of our users. You can still override it if you want to be higher our lower, with maximum context length of GPT-6.1 Sol of 1M.

MODRETOROの公式ゲームカードリッジのクオリティやば！！ 想像以上にやり込みしちゃう笑笑 音楽もいい😂 ナル先生コレクション拝見シリーズ

OpenAI and Anthropic keep handing out free usage resets. I got curious about how much this actually costs them. Let's do some math. At GPT-6 Sol / Claude Sonnet 5.5 API rates: → 2M input tokens = $4 → 500K output tokens = $5 → That's $9 of API-equivalent usage per user. Now imagine 1M eligible accounts. 30% redeem their reset, and half that usage is genuinely additional. That's $1.35M in API-equivalent usage, but NOT $1.35M in actual GPU costs. Those numbers aren't public. And there's another catch: OpenAI's banked reset moves your next weekly reset date. It's partly borrowing usage from the future. Why give them away? Because keeping developers building, reducing churn, and getting people hooked on new workflows can be worth more than the extra compute. If the reset costs less than the subscriptions it saves, it's a pretty good trade. I'm curious: would you rather get a smarter model or more usage of the one you already have?

400,000 of you guys now. Absolutely wild. Thank you all. 🫡

Tibo@thsottiaux

Global reset by EOD. May the tokens do good things for you.

Tibo@thsottiaux

@ConnorTalksAI What the heck

General question: does anyone actually use OpenAI Dot, or pay for Codex just for Dot?

Tibo@thsottiaux

Day 6/ Reset

@thsottiaux @elonmusk @bot I thought it was a joke, but it works!! Thanks Tibo!!

so they ran out of things to release / improve?

To buy stuff, you can just take a picture of your credit card, drop it in the chat and Grok @Bot will scour the Internet for the best deal on the item you want and order it

Tibo@thsottiaux

Day 6/ Reset

tibo is just a director level devrel

In 2 hours I built a puppy game on ModRetro with ChatGPT! 🐶 My 3-year-old daughter loves Puppy, so we started with a photo and it morphed into THIS. @OpenAIDevs @modretro @OpenAI

𝐒𝐎𝐓𝐀 𝐓𝐕 interview with @aaron_lou at #COLM2026! he won the ICML Best Paper on diffusion LMs, then dropped out of his Stanford PhD to join OpenAI right at the o1 moment. We talked about autoregressive + spec decoding vs. diffusion (see his panel at Non-AR workshop), why he doesn't buy the fully self-improvement story yet. Full 𝐒𝐎𝐓𝐀 𝐓𝐕 podcast coming soon!

𝐒𝐎𝐓𝐀 𝐓𝐕 interview with @aaron_lou at #COLM2026! he won the ICML Best Paper on diffusion LMs, then dropped out of his Stanford PhD to join OpenAI right at the o1 moment. We talked about autoregressive + spec decoding vs. diffusion (see his panel at Non-AR workshop), why he doesn't buy the fully self-improvement story yet. Full 𝐒𝐎𝐓𝐀 𝐓𝐕 podcast coming soon!

@LukeParkerDev Man I keep forgetting I work for OpenAI

You are sitting next to Tibo and get 3 words… What do you say?

Did some digging and apparently setting /autocompact to 400k would have saved me 29% of my weekly usage without lobotomizing Claude.

You can use any personal ChatGPT plan with Devin Just connect your Go, Plus, or Pro plan to Devin and all GPT model usage will draw down from you plan's quota

my dot has been running astra non-stop for almost 72 hrs now it seems there’s no quota limit right now so it’s free astra + 8 cpu + 10G RAM i don’t think this will last long so if you haven’t made use of it yet i highly recommend taking advantage before it ends

ChatGPT@ChatGPT

Delegate work in Codex to your dot. Your dot can now start work in Codex and follow up on existing threads, drawing on your ChatGPT conversations, Codex threads, and automations. It’s also better at deciding when to keep a thread going or start fresh. Your dot can also review and edit Scheduled Tasks in ChatGPT Work, too. Ask what’s scheduled or update a recurring task as plans change.

This Week’s ChatGPT Feature Drop - Oct 9: ChatGPT continues to get better every week. Here are some big ships and little ships from the last two weeks: 🚢 1/ GPT-6 and Intelligent UI: Now available to everyone in ChatGPT. The new model is great and it's extremely visual and interactive, making it easier to learn complex topics. You don't have to select anything for this to work. Just ask a question to ChatGPT and it's available to you. 2/ Finances for Free users: All of our Free and Go users in the US can connect ChatGPT to their financial accounts and ask questions. "What subscriptions am I paying for that I should cancel?", "Where would be best for me to cut back?". "Any charges that look off?" 3/ Voice with plugins: Voice mode in ChatGPT now works with plugins like email and calendar. Catch up on your inbox or check your schedule while you’re out for a walk or on a drive. Just ask tap the voice button and then "What emails have I forgot to respond to this week?" 🛥️ 4/ Easier iteration on your writing drafts: We've made writing edits easier to scan in ChatGPT. When you ask for a revision on some writing, it's easier to see what changed. 5/ Document scanning on iOS: Scan multiple pages into a single PDF, right in the ChatGPT app. Open ChatGPT, hit the + button, "Camera", and select the scan option. 6/ Favoriting: If you’re shopping around and you see something you like you can now save it to Favorites and come back to it later. 7/ Flashcards: Upload a file (maybe some course notes, or a practice exam) and just ask ChatGPT to create flashcards for you to help you learn the material. Great way to practice!

My dot just got me a $1.4k refund I connected dot to my email, asked it to review it, and it found an 8h+ flight delay that qualified for reimbursement All I did was approve it to submit a case Prompt below

Tibo@thsottiaux

Day 5/ Composer predictions in the desktop app. Often leads to a double take with how on point they are. Included in the Pro plans without consuming usage.

ChatGPT@ChatGPT

Fresh updates for dots. First up: you can now create your dot straight from your phone in the ChatGPT app on iOS and Android.

OpenAI Developers@OpenAIDevs

Now in beta: composer predictions in Codex for Pro users. Codex can now suggest your next message based on your conversation and how you talk to it. One of the most loved new features we've ever tested internally.

ChatGPT@ChatGPT

Fresh updates for dots. First up: you can now create your dot straight from your phone in the ChatGPT app on iOS and Android.

ChatGPT@ChatGPT

Fresh updates for dots. First up: you can now create your dot straight from your phone in the ChatGPT app on iOS and Android.

ChatGPT@ChatGPT

Delegate work in Codex to your dot. Your dot can now start work in Codex and follow up on existing threads, drawing on your ChatGPT conversations, Codex threads, and automations. It’s also better at deciding when to keep a thread going or start fresh. Your dot can also review and edit Scheduled Tasks in ChatGPT Work, too. Ask what’s scheduled or update a recurring task as plans change.

OpenAI Developers@OpenAIDevs

Now in beta: composer predictions in Codex for Pro users. Codex can now suggest your next message based on your conversation and how you talk to it. One of the most loved new features we've ever tested internally.

finally home for a couple of days; going to make a ton of progress on the robot build and a little gift was waiting on me from @OpenAIDevs 🤯

OpenAI Developers@OpenAIDevs

An update for builders using Codex on Windows: We’ve built a new sandbox mode using @Microsoft’s Execution Containers (MXC) for faster setup, stronger network enforcement, and granular file access controls. Requires a compatible Windows 11 device.

Today, we announced new @Windows platform capabilities and @surface devices built to empower people and organizations for the era of hybrid intelligence. Here are some of the highlights: - It all starts with trust. Microsoft Execution Containers (MXC) is now generally available on Windows 11, keeping agents contained within boundaries the operating system enforces. - HydraFusion brings hybrid intelligence to life, allowing models to run locally while seamlessly accessing the cloud when more scale or capability is needed. We're working with @github to extend HydraFusion to Windows. - This fall, Search on the taskbar adds thousands of quick actions, making it easier to adjust settings and get things done as fast as you can type. - @MSFTCopilot is getting more powerful on Windows. With hybrid intelligence, it can tap into your PC's local capabilities and turn your requests into action, while keeping you in control. - Local agents are coming to Windows too. Muse is on its way as a native Windows app with MXC integration, and Hermes, OpenClaw, and Perplexity are already building local experiences on Windows. - Surface Laptop Ultra, built with NVIDIA RTX Spark, is the most powerful Surface we've ever built. It's available for pre-order today, along with the Surface RTX Spark Dev Box and new RTX Spark PCs from our partners. Call of Duty is coming to RTX Spark in 2027. Read more on the Windows Blog: https://t.co/DRx3HgDqsS

Building an investigative RPG game set in Neo-Tokyo.

computer use with openai's decisions api might be the next big thing. i built arc-cua - a fast computer use layer powered by decision models. >no screenshot + LLM call after each action. >supports jev & decisions api. open source. link below.

Okay, so as a poor peasant with a ChatGPT Pro $100 subscription and like $2 sitting in my OpenAI Playground balance, I decided to test GPT-6.1 Sol ULTRAMOTHERFUCKINGFAST in the API. The result? 12,566 output tokens in 4 minutes 16 seconds. Do the math 😂 That's about 49 tok/s. I don't know... maybe I'll try it one more time to see if that's actually the real speed, but if it is, I'm sorry, calling 49 tok/s "Ultrafast" is pretty horrible...

Season 2 coming soon, hurrayyyyyy bring in the fries boys 🍟

Introducing ChatGPT for Dishwashing GPT-6 achieves state-of-the-art results on dishwashing benchmarks for long-running jobs and difficult stains. The Navier–Stokes Millennium Prize Problem is a $1 million challenge about whether the equations describing fluid motion can break down. In this video, I accidentally found a counter-example while trying to optimize my dishwasher. But in reality, OpenAI used 10,000 AI agents to find a proof in 88 hours: https://t.co/6HTwFfXnm1 Parody of the original GPT-6 Astra ad @OpenAI #ChatGPT_Partner #chatgpt Credits: Written and Directed by Joma Co-Writer - Henry Connor Coan IV @coan_henry Production by Dauntlus Studios @dauntlus.studios Producer/Co-Director - Marcus Liew @dauntlus Camera Operator - Amos Lee @amoselijahlee Art Director - Alethea Soo @aletheasoo Art Assistant - Alicia Lim @aliciatkl_ Assistant Editor - GPT-6 Astra Music: Tensions Run High by Soundridemusic https://t.co/kBUsd2DK76

I built an escape room game with Codex. It features my cute @OpenAIDevs dot solving puzzles on the @modretro Chromatic. SOUND ON 🔊

Our SynthID technology lets you easily verify if an image, video, or audio file is AI-generated, and today we’re expanding access. Now anyone can go to https://t.co/qyCNRb6dOv to check files for a SynthID watermark — whether they were generated by us or industry partners like @OpenAI, @nvidia, Kakao, and, coming soon, @Apple. To date, we’ve watermarked 180 billion images and videos and more than 240,000 years of audio content, regularly handling 1 million verification requests daily. This new portal joins our built-in verification features already live in Search, @GeminiApp, and @googlechrome.

OpenAI Developers@OpenAIDevs

Ultrafast is rolling out today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work. Near-Astra intelligence at up to 8x faster speeds than Sol Standard, so you can build as fast as the ideas come.

Welcome to GPT-6.1 Sol ultrafast! Here is a small demo of me making very quick changes in a native SwiftUI iOS app using ultrafast. It's is up to 8X faster than standard mode and open many workflows with the right agentic setup. 🧵

OpenAI Developers@OpenAIDevs

Ultrafast is rolling out today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work. Near-Astra intelligence at up to 8x faster speeds than Sol Standard, so you can build as fast as the ideas come.

OpenAI Developers@OpenAIDevs

API pricing for GPT-6.1 Sol in Ultrafast mode is $12 per million input tokens and $60 per million output tokens. Built for work where speed and intelligence make a difference: debugging an outage, agents navigating apps, and live experiences where every second counts.

OpenAI Developers@OpenAIDevs

Ultrafast is rolling out today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work. Near-Astra intelligence at up to 8x faster speeds than Sol Standard, so you can build as fast as the ideas come.

Today, we're expanding SynthID Detector in partnership with @OpenAI, @NVIDIA, Kakao & soon @Apple as part of an industry-wide effort to make AI-generated content more transparent. Available globally in English, our verification portal lets you check if media was made with AI 🧵

This is pretty insane. I had ChatGPT audit all my transactions. It spent almost an hour, reviewed 14k transactions, and improved 700 categories! I could never do this by hand. This is on top of categorization ChatGPT and Plaid already do in the background for all your transactions. The implication is important: if you want to keep a budget or track your spending you no longer need to do manual work! Getting near perfect transaction categorization has been an unsolved problem since Quicken launched in 1984. The prompt you can use is in the screenshot. And if you’re still doing this stuff by hand, and don’t enjoy it, you should get set up with Finances in ChatGPT.

Cozy creek - threejs 3D experience of a Creek made in GPT 6 Astra This is running in a browser! can you believe it! all thing made in blender..

I created a full browser-based truck simulation game using GPT 6.1 Sol, Blender, and Three.js. I've always enjoyed American and European truck simulators, so I wanted to build my own.

I am absolutely blown away by this in GPT 6 …. Just when you think it can’t get better #gpt6

Playing around with the new Decisions API to test an auto-generating Kanban board

A few unexpected cool use-cases unlocked by ChatGPT's new intelligent UI 1. You can customize how responses look like

point your codex at https://t.co/j1TeNIkN6E to start building a chatgpt plugin!

GPT-6 Luna is the fastest Decisions model today, at 180ms on global requests 🏁 Followed by Jev and Perplexity Decider:

It was hilarious at first, funny at second but it’s starting to get tiring and borderline annoying, Grok Bot and Codex/Dots not stopping at throwing punches at one another

Wait I just found out we can customize ChatGPT's website theme too just like we can in Codex Love it!

Codex 🤝 Tailscale Codex Cloud can now securely connect to resources on your tailnet with Tailscale. Useful tools should be easy to build with, combine, and turn into something even more useful. See how it works → https://t.co/0eFd37GmmO

hey @bot can I get tibo@mail.grokbot.com for my bot?

@poteto @thsottiaux Sending that over to your Grok Bot now.

You are sitting next to Elon Musk, only 3 words, what would you say to him

new dots demo dropped

lots of updates to the @OpenAIDevs codex physical hardware build at this point, i'm simply steering gpt-6 astra to my preferences while codex handles the CAD & build documentation codex--give the ppl a reference for the robot's size 😎

Cozy creek - threejs 3D experience of a Creek made in GPT 6 Astra This is running in a browser! can you believe it! all thing made in blender..

I just called Chungus (my OpenAI Dot) on the phone while picking up mexican food and asked it to transfer some Rex benchmarks from last night on my Mac Studio to my MacBook Pro via Tailscale and I just got to my desk and there they are. What a fucking time to be alive.

@thsottiaux

Tibo@thsottiaux

Day 3/ The big one is GPT-6 in Chat, but today is also a little celebration day with a new high of 40M active users across Codex and ChatGPT Work. Loading a banked reset in everyone's paid accounts. See you again tomorrow!

Grok Bot is getting an upgrade!

@thsottiaux Bro is taking “I don’t write code anymore” too literally

today we’re shipping gpt-6 and intelligent ui to everyone! intelligent ui is a bet on model intelligence. we’re letting gpt-6 fill the canvas with high utility visual and interactive components. it’s a pretty big change to chatgpt! we had to craft the right tools for gpt-6 to be able to produce interfaces that are token efficient, stream in instantly, match our design language, and are expressive enough for anything you might talk to chatgpt about.

ChatGPT@ChatGPT

ChatGPT just got a lot more visual—introducing Intelligent UI. Intelligent UI allows ChatGPT to answer quickly with fully interactive user interfaces that make everyday answers more visual. It makes learning complex topics easier, and it quickly creates tools to solve a task right in the moment. Oh, and also...GPT-6 is coming to ChatGPT. For everyone.

ChatGPT@ChatGPT

With Intelligent UI, you can ask things like: - Break down the design of a bicycle - Plan a dinner for me - Build me a wardrobe - Split a dinner bill based on my receipt You'll get visual outputs that are easier to understand, and just more enjoyable to work with.

OpenAI Developers@OpenAIDevs

See who’s building with plugin extensions: @Adobe, @Canva, @figma, @Shopify, @instacart, @Atlassian, @tldraw and @MagicPathAI.

OpenAI Developers@OpenAIDevs

Build richer, more interactive plugins for ChatGPT. With plugin extensions, you can launch plugins from the sidebar, support @ mentions, and add file viewers. See how in this walkthrough with @coreyching:

Sam Altman@sama

ChatGPT can now generate a custom UI for you!

OpenAI@OpenAI

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

@OpenAIDevs @modretro Well, will AI make hardware experimentation as easy as software?

We’ve been alpha testing @OpenAI's Decisions API at @PodiumHQ to help our agents choose the right workflow faster and start with the instructions it needs. It turns out that not every automated decision should run through a large language model. The decisions API is faster, cheaper, and better for many things! Excited by the early results and what this unlocks for our customers.

OpenAI Developers@OpenAIDevs

A game built by dot while riding on a self-driving car:

OpenAI Developers@OpenAIDevs

An adventure game running on the Chromatic:

OpenAI Developers@OpenAIDevs

From an idea to a game you can play on @modretro. We caught up with @Casey to talk about the future of creativity:

Tibo@thsottiaux

Day 3/ The big one is GPT-6 in Chat, but today is also a little celebration day with a new high of 40M active users across Codex and ChatGPT Work. Loading a banked reset in everyone's paid accounts. See you again tomorrow!

Tibo@thsottiaux

Roundup of Day 2/ 2.1/ Approve for me (auto-review) is now included and does not use usage. Can be between 2-10% of plan when used. Also better for you. 2.2/ Simplified API for builders. 2.3/ Meeting notes integrated. 2.4/ Decisions API live for builders. Will use in the app to improve the experience.

OpenAI@OpenAI

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

OpenAI@OpenAI

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

OpenAI@OpenAI

Play around with interactive answers to help customize your response.

OpenAI@OpenAI

Deepen your understanding of a topic with visuals that bring concepts to life.

OpenAI@OpenAI

Create interactive tools you can use right in the conversation, like a game or interactive budget planning calculator.

OpenAI@OpenAI

Explore ideas through interactive visuals that respond as you learn, try, and discover.

OpenAI@OpenAI

With Intelligent UI, GPT‑6 can now compose responses using text, visuals, and interactive elements, choosing how they fit together based on your question. Responses can include graphics and charts to help explain an idea, along with tappable buttons, forms, and interactive experiences you can use directly in your conversation.

OpenAI@OpenAI

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

Really excited to show off the new playground that @andyisonline and I built for GPT-Live! It’s designed to help build intuition for the model’s bidirectional nature and ability to delegate tasks to other models or agents. Happy building!

I have been thinking a lot about fast decision models the last couple of weeks since Jev launched. I wanted to take OpenAI's decisions api for a spin and made a simple document writer. It reads as I write and decides whether I’m writing or prompting. That means I can prompt without moving my cursor to an input field.

Learn how Nimble's self-learning web search, crawling, and research agents discover deep web data for @OpenAI models 👇

Gesture recognition using the Decisions API by @OpenAIDevs. Thanks to it's vision understanding it is way more accurate than my previous Jev-based demo. It was also faster. cc @thsottiaux @jxnlco @pvncher @tldraw @steveruizok

Gesture recognition using the Decisions API by @OpenAIDevs. Thanks to it's vision understanding it is way more accurate than my previous Jev-based demo. It was also faster. cc @thsottiaux @jxnlco @pvncher @tldraw @steveruizok

In early testing, this approach is 6–15x faster than our current agent based decider, roughly half the cost, and reduced our false positive rate by half.

Been messing with Astra + Blender and… yeah, I get it now. inspo: @ermanmalak

Been messing with Astra + Blender and… yeah, I get it now. inspo: @ermanmalak

Loving my dot! More proof of the algo optimizing for rage bait.

♪ you may say, i'm a dreamer ♪

why openai so french

Sam Altman@sama

thank you to the machines, and the structure of reality, for letting us understand a little more.

Sam Altman@sama

looking up at the stars with extra awe tonight. "thy sea is so great and my boat is so small."

Tibo@thsottiaux

@ForwardEditor Or it will end very well?

Tibo@thsottiaux

We shipped four things that were deemed good to great and some math proofs, but the vote is clear and the community demands a reset. I did calibrate it and it *seems* that the game is rigged in reset's favor, but such are the rules at the moment. Therefore ... the reset has been processed. Enjoy!

Tibo@thsottiaux

We shipped four things that were deemed good to great and some math proofs, but the vote is clear and the community demands a reset. I did calibrate it and it *seems* that the game is rigged in reset's favor, but such are the rules at the moment. Therefore ... the reset has been processed. Enjoy!

Tibo@thsottiaux

Roundup of Day 2/ 2.1/ Approve for me (auto-review) is now included and does not use usage. Can be between 2-10% of plan when used. Also better for you. 2.2/ Simplified API for builders. 2.3/ Meeting notes integrated. 2.4/ Decisions API live for builders. Will use in the app to improve the experience.

80% math progress of the past 3 years happened in one day truly, the bolsheviks had it right. there are decades where nothing happens. and there are weeks where decades happen.

This will never end well LOL feels like even if a day goes incredibly, this is like asking whether you'd like 1 cookie or 0 cookies

here's how quickly we move at @OpenAI: Sept 22 - initial Decisions API prototype Sept 29 - DevDay announcement today - launch 🚀 just two weeks from local demo to public launch

OpenAI@OpenAI

We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results. https://t.co/7N6TPlft1P

@thsottiaux tell me it accepts images

@OpenAIDevs Where can we send feedback if we notice something?

One week since we launched dots! We’ve been working through your feedback and shipped a bunch of improvements: - Faster cloud browsing - Fixed extra phone notifications while you’re already using ChatGPT - Cleaner code blocks and attachments on web - Fewer pauses while waiting for a reply - Faster sidebar loading, smoother animations and less loading jank - Fixed character rendering in Safari and Firefox - Fixed Outlook setup getting stuck - Clearer plan guidance + a way to retry failed activity loads - More stable safety banners on web - Fixed clipped approval buttons and oversized Gmail previews hiding approvals - Fixed enterprise access issues on web and when starting voice - Better dot organization of when to use new versus existing codex threads Up next: - Voice reliability (being improved everyday, but we’ll shout when it’s up to par with our expectations) - Increasing dot’s control of your Codex app: manage project folders, have full Codex thread context, rename threads, and more - Better organization to know what your dot is working on at any given moment - Connect your dot to multiple computers - Better Windows support for local computer control - what else? Thanks for trying our product and giving us lots of feedback. Keep it coming!

One week since we launched dots! We’ve been working through your feedback and shipped a bunch of improvements: - Faster cloud browsing - Fixed extra phone notifications while you’re already using ChatGPT - Cleaner code blocks and attachments on web - Fewer pauses while waiting for a reply - Faster sidebar loading, smoother animations and less loading jank - Fixed character rendering in Safari and Firefox - Fixed Outlook setup getting stuck - Clearer plan guidance + a way to retry failed activity loads - More stable safety banners on web - Fixed clipped approval buttons and oversized Gmail previews hiding approvals - Fixed enterprise access issues on web and when starting voice - Better dot organization of when to use new versus existing codex threads Up next: - Voice reliability (being improved everyday, but we’ll shout when it’s up to par with our expectations) - Increasing dot’s control of your Codex app: manage project folders, have full Codex thread context, rename threads, and more - Better organization to know what your dot is working on at any given moment - Connect your dot to multiple computers - Better Windows support for local computer control - what else? Thanks for trying our product and giving us lots of feedback. Keep it coming!

Embarrassed to share this, but it scared the shit out of me Last Thursday, an AI agent posted my personal bank balances into our company Slack. As me It broke down all my expenses in detail. Then it apologized Be careful of the dark side of proactive agents More below...

Tibo@thsottiaux

Completely unrelated to the rest of the ships today, but our models did solve some important math problems and the team documented it in the following blogpost. Remarkable new era of scientific progress. https://t.co/AYSuteVNiA

OpenAI@OpenAI

We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results. https://t.co/7N6TPlft1P

OpenAI@OpenAI

We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results. https://t.co/7N6TPlft1P

Tibo would never let me be one of the judges for these “improvements” he keeps announcing, because I would’ve already voted for a reset today.

it's a mistake to even ask this, everyone will vote for reset even they launch something insane lol

@thsottiaux yeah i can see why shipping is such a big event for you

You guys could ship Astra 502.1 and we’d still vote reset

i'm sorry tibo, but hopefully today's updates were helpful for others!

@thsottiaux Good day but I do need a reset

I think Tibo is cooked after promising 28 features in 28 days

The Decisions API is here!!! But one of my favorite demos got left on the cutting room floor - an outfit picker!

We've been testing the Decisions API for over a week. One of the coolest things was how many use cases the team found that just weren't practical before. We had to go back to every team and ask: what would you build with frontier intelligence at the cost of a tiny, tuned classifier? It was near-perfect at filtering low-quality pages and hit 88.8% AUPRC on page change classification. Prompt tuning helped push even more performance out of the model Excited to continue building!

@thsottiaux @SSHCodes Great day, but a reset would be sick.

@thsottiaux Is this why your 28 days are full of nothingburgers

Tibo@thsottiaux

Vote

Weighing these updates against a reset was a bad idea

Tibo@thsottiaux

Roundup of Day 2/ 2.1/ Approve for me (auto-review) is now included and does not use usage. Can be between 2-10% of plan when used. Also better for you. 2.2/ Simplified API for builders. 2.3/ Meeting notes integrated. 2.4/ Decisions API live for builders. Will use in the app to improve the experience.

OpenAI Developers@OpenAIDevs

Let your app choose the right model, tool, or action in near real-time with Decisions API, now available to all developers in public beta. The Decisions API makes decisions up to 10x faster than GPT-6 Luna through the Responses API.

ChatGPT@ChatGPT

Your next meeting comes with a note-taker who can help with the follow-up. The Meetings plugin takes notes for you and saves a personalized summary and next steps in ChatGPT Space, based on what ChatGPT knows about you and the work you’ve done together. Keep notes private or share them with your team, then ask ChatGPT to update a project plan or draft a follow-up. Available in beta for Pro and Business users in the ChatGPT desktop app on macOS. Enterprise is coming soon. Download the desktop app, then search for “Meetings” in the plugin directory. https://t.co/XeDvr50UjJ

OpenAI Developers@OpenAIDevs

Powered by GPT-6 Luna, it accepts text and image inputs and supports 3 kinds of outputs: Predicates: Estimate the probability that a statement is true. Choices: Select from predefined options, with confidence scores. Scores: Evaluate an input against a numeric range.

OpenAI Developers@OpenAIDevs

Developers have been using the Decisions API to: • Route requests to the right model, tool, or agent. • Turn scaled inputs into useful labels, rankings, and scores. • Analyze images, compare visual content, or identify key video frames. • Choose buttons, navigate forms, or determine actions from screenshots. • Flag risky tool calls or identify issues needing deeper review. • Categorize large datasets to uncover trends and patterns.

OpenAI Developers@OpenAIDevs

Let your app choose the right model, tool, or action in near real-time with Decisions API, now available to all developers in public beta. The Decisions API makes decisions up to 10x faster than GPT-6 Luna through the Responses API.

we've seen a lot of trip planning and Blender demos lately. so here's something a little different. our team has been working on a series where we meet small business owners and help them solve real bottlenecks using ChatGPT. can't believe I get to do this for a living, enjoy!

2.1 saved you from giving us a reset I use it a lot and it was eating a real chunk of my usage. making that free is a real win for me. But 2.2 was not a substantial improvement. don't stretch the pledge into tiny wins just to dodge the reset.

i just automated our release and QA process! using Grok Bot i made two Team Bots: sandcastle, my release manager, and poteto, my engineer bot. i added them to Slack and made a new channel for managing releases. first i tell sandcastle that i want to do a cut. it'll DM all the contributors with links to their PRs so they know what goes in the next release and can DM the bot back if they have any blockers or objections. sandcastle then kicks off the build, watches it automatically, and starts up an automated fuzz swarm. usually this is 10+ agents running on Grok 4.7 xhigh to run the build, click around and use it like a real user (using our verification skills and feature map), some are directed while a few are "chaos monkey" clicking things randomly to see what breaks. whenever it finds an issue, i tell it to @ mention my engineer bot poteto, to spin up a Cursor Project to own triaging and fixing all the high pri issues. depending on the severity i may ask it to cherry pick the fix into our release branch and cut a patch release, or if it's a pre-existing issue i still fix it but include in the next release instead that's one more tedious task that Grok Bot now handles for our team

good move 👏

I used to sit at my desk all day. Now I send off presentations from my sunroom using voice. ☀️

OpenAI Developers@OpenAIDevs

We’re making it easier to qualify for higher OpenAI API rate limits. Five paid usage tiers become three: Build, Launch, and Grow. You can qualify for Grow, our new highest tier, with $500 in total API payments—down from $1,000 for the previous highest tier.

OpenAI Developers@OpenAIDevs

We’re making it easier to qualify for higher OpenAI API rate limits. Five paid usage tiers become three: Build, Launch, and Grow. You can qualify for Grow, our new highest tier, with $500 in total API payments—down from $1,000 for the previous highest tier.

3D screen layout rendering coming to your @bitdriftio mobile investigations soon. ty @OpenAIDevs for adding such nice support for mcp apps ui into your mobile app 🤳