Resources:
Join the discussion on LinkedIn: Got something to say? Let us know on LinkedIn and network with other AI leaders
Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup
Connect with Jordan Wilson: LinkedIn Profile
Start Here Series in our Inner Circle Community: Join for free access
7 AI Features You Can Use Right Now: ChatGPT Tasks, Claude Design, Codex Record and Replay, and More
A busy week in AI just handed business users a stack of genuinely usable upgrades. From OpenAI bringing back scheduled tasks to Anthropic overhauling Claude Design, here is a breakdown of the 7 features worth your time right now.
ChatGPT Tasks Are Back (And Better Than Before)
OpenAI is sunsetting Pulse, its proactive daily summary feed, and replacing it with a revamped ChatGPT Tasks. Tasks were actually the original feature before Pulse, but on certain paid plans including ChatGPT Pro, Tasks had been quietly removed in favor of Pulse. Now they're coming back, and rolling out to all paid plans.
The updated Tasks feature lets ChatGPT perform work proactively in the background, including setting reminders, managing recurring work, and monitoring connected apps on a schedule. A new dedicated scheduled page lives in the ChatGPT sidebar, where users can view, pause, resume, edit, and delete all active tasks in one place.
One meaningfully new capability: you don't have to specify a check frequency. If you want ChatGPT to monitor a competitor's product page and alert you when something changes, you can just say "tell me when it's updated" rather than setting a daily schedule. The system figures out when to check. The feature also plugs into connected apps like Gmail, Google Calendar, and Google Drive, making it possible to run a daily briefing that pulls from live data every morning. Currently, users can create up to 10 tasks at a time. The feature is available on web, iOS, and Android. No word yet on free tier access, though there are indications it may come.
Claude Design Gets a Major Upgrade
Claude Design launched with solid output quality but a frustrating workflow. Everything defaulted to the same cookie-cutter aesthetic and the tool lived in its own isolated tab with limited integration into the rest of the Claude ecosystem. The Claude Design June update changes most of that.
The headline addition is a true WYSIWYG editor, letting users click directly on elements and edit them in place, similar to working in Canva or Wix. New rich layout controls let users drag, resize, and align elements, and the team says hundreds of stability fixes are included to make the editor hold up under real use.
Equally significant is design system import support. Users can now bring in their own component libraries from a GitHub repo, design file, or raw upload. Claude then builds against those components instead of inventing its own buttons and spacing. That imported system is the same component library Claude Code uses, which creates a genuine round trip from design prototype to production code without respeccing the design.
New export options include PDF and PowerPoint, and direct integrations now exist with Canva, Gamma, Lovable, Miro, Replite, Vercel, and Wix. Claude Design is available in beta on paid Claude plans (Pro, Max, Team, Enterprise) at claude.ai/design and now has a home in the Claude desktop app sidebar.
Google Vids AI Avatars Expand
Google Vids got a meaningful upgrade to its AI avatars, combining Gemini 3.1 Flash text-to-speech with the latest Veo 3.1 video model. The result is more realistic, expressive avatars backed by expanded language support.
The avatar library grew from 23 to 53 default presets, spanning photorealistic, 3D cartoon, and graphic novel styles. More importantly, video generation now runs without a duration limit, up from the previous 8-second cap. Custom avatars can now be directed via text prompt to walk, talk, and interact with objects in a scene, including picking up props and moving around the frame. That removes a lot of the robotic stiffness that has made AI avatars feel gimmicky.
A new Google Slides integration lets users convert presentations into videos with a few clicks. Anyone with a US Google account can access the feature at no additional cost, with up to 10 Veo generations per month on the free tier and higher quotas on paid Workspace plans.
OpenRouter Fusion: Multi-Model Synthesis Made Easy
OpenRouter Fusion lets users run a single prompt through multiple AI models in parallel, then uses a judge model to synthesize a single best output. The quality preset pairs two frontier models together; the budget preset runs a combination of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro.
Fusion launched as a public experiment a few months ago and has now been fully integrated into OpenRouter's API, with a 128K context token window. Users can try it at openrouter.ai/fusion with quality, budget, and custom model options. Running multiple models in sequence and having a judge evaluate the output is something many experienced users already do manually. Fusion automates that process without requiring any custom routing infrastructure.
Claude Code Artifacts: Live Shareable Pages from Code Sessions
Claude Code Artifacts brings the familiar artifacts concept into Claude Code, but specifically for Team and Enterprise plan users. The feature lets Claude Code capture the progress and context of a session into a live, shareable visual page.
Use cases include PR walkthroughs, system explainers, dashboards, and release checklists that update themselves as the session advances. Artifacts publish to a persistent URL with version history and an org-level gallery. Teammates see the same live view with the same context and get updates in real time as Claude Code publishes changes. This is comparable to Codex Sites from OpenAI, though the Claude Code version is more engineering-oriented. Product managers who want to track what actually shipped without asking engineers directly are also a natural fit.
GLM 5.2 from ZAI: The Strongest Open Weights Model Right Now
GLM 5.2 from ZAI is a 753 billion parameter mixture of experts model released under an MIT license, with weights available on Hugging Face. It is currently ranked as the leading open weights model on the Artificial Analysis intelligence index, scoring higher than Gemini 3.1 Pro.
It is text-only and not multimodal, but comes with a 1 million token context window, a significant increase from the 200K offered by the previous version. ZAI positioned it for long-horizon autonomous coding and engineering tasks, and the lab claims it beats GPT-4.5 on several long-horizon coding benchmarks at roughly one sixth of the cost. On front-end coding specifically, it currently ranks first in the arena, ahead of any available model.
Because it carries a fully commercial MIT license, it can be downloaded, self-hosted, fine-tuned, and deployed air-gapped. For enterprises focused on token efficiency and vendor independence, that combination is hard to ignore. The model is available via ZAI's website and API.
OpenAI Codex Record and Replay: Turn Any Workflow Into a Skill
Codex Record and Replay is the most immediately practical feature in this batch. The idea is simple: click a button, do your work, and Codex watches. It learns the steps you took and packages them into a reusable skill or automation that can be triggered at any time.
OpenAI's framing is clear: this is for workflows that are repetitive, dependent on personal preferences, or easier to demonstrate than to describe in a prompt. Examples include filing expenses, booking a parking space, creating a correctly configured issue, publishing a video, or downloading a recurring report. Codex packages the recorded pattern into a skill compatible with computer use, browser actions, and connected plugins.
This puts Codex into direct competition with traditional RPA (robotic process automation) tools like UiPath and Automation Anywhere, but with a substantially lower setup bar. Users who already have projects, apps, and workflows set up inside Codex benefit immediately because the recorded skill inherits that existing context.
Currently available only on Mac OS, and not yet in the EU, UK, or Switzerland. Access requires a paid ChatGPT plan. The Codex team has indicated that voice narration during recording may come in a future update.
Topics Covered in This Episode:
- ChatGPT Pulse Sunsetting and Tasks Comeback
2. ChatGPT Scheduled Tasks Features and Access Tiers
3. Claude Design June Update Overview
4. WYSIWYG Editing and Design System Imports in Claude Design
5. Claude Design Export Options and Third-Party Integrations
6. Google Vids AI Avatars Upgrade with Veo 3.1
7. OpenRouter Fusion Multi-Model Synthesis Feature
8. Claude Code Artifacts for Team and Enterprise Plans
9. GLM 5.2 from ZAI Open Weights Model Overview
10. GLM 5.2 Benchmarks and Enterprise Use Cases
11. OpenAI Codex Record and Replay Feature Explained
12. Codex Record and Replay vs. Traditional RPA Tools
Episode Transcript
Jordan Wilson [00:00:02]:
ChatGPT tasks are back like they never left, even though they did, but they also kind of didn't. Claude Design went from a slow and expensive party trick to a creative juggernaut in training. Claude Code is getting in on the artifacts game, and Codex can now just watch you work and then repeat back all the tasks that you just did.
Oh, we got a new model that scores higher than Gemini 3.1 Pro. And it's not from Google, Anthropic, or OpenAI. And it's open weights. What? Yeah.
Jordan Wilson [00:00:42]:
Crazy week of new AI features that you can use today. So while we were all paying so close attention to what was happening with Anthropic and the US government and Fable V, well, everyone just released a ton of new AI features that you can start using today. But unless you were spending hours every single day keeping up with it like I do, you probably missed a lot of these new updates.
That's why we have our Friday Features show that brings you a bunch of new AI upgrades that you don't have to wait for. No wait lists. No coming soon. Go use them now. So welcome to Everyday AI.
Jordan Wilson [00:01:26]:
But let's talk about what you're going to learn on today's show. You're going to learn why ChatGPT getting rid of one feature was actually a big upgrade. You're going to know the simple Claude Code design upgrade that will make an instant difference. And you'll know if one of the biggest names in tech will make a dent in the AI avatar space.
All right, let's start. Welcome to Everyday AI. My name is Jordan Wilson. We do this thing for you. It's a daily live stream, podcast, and free daily newsletter helping everyday business leaders like you and me not just keep up with all of these updates, but I tell you what to use, what not, and how to use it to grow your company and career. So it starts here with the unedited, unscripted live stream podcast, but make sure to go to our website at youreverydayai.com and we're going to give you not just the recap of today's show with all the important links so you can go try all these new features, but also we're going to give you all of the other AI news and updates you need to know to be the smartest person in AI at your company.
Jordan Wilson [00:02:32]:
All right, let's get straight into it. A ton going on. Let's start with ChatGPT Tasks. So this is technically something getting taken away because this announcement is actually ChatGPT's Pulse going away and being replaced by ChatGPT Tasks. Let me explain.
So before, ChatGPT Tasks was first, right? And then OpenAI kind of demoed and slowly rolled out this thing called Pulse, which was this proactive kind of agent that gave you kind of like a nice little web page every single day based on your interests, different things that you were using ChatGPT for, et cetera.
Jordan Wilson [00:03:20]:
Personally, I didn't like Pulse. I thought it was hard to steer. I didn't find a lot of value. But the thing that I actually disliked was that on my ChatGPT Pro plan, at least, I couldn't use Scheduled Tasks anymore because Scheduled Tasks were the precursor to Pulse. And then they kind of got rid of Tasks on the Pro plan. And I'm like, oh my gosh, I need my Tasks back. And well, now Tasks are back, but they're just back because Pulse is gone. But now Tasks are rolling out to everyone. Confusing, I know.
All right, so here's what it is. ChatGPT can now perform tasks proactively in the background, including setting reminders, managing recurring work, and monitoring connected apps on the web on a schedule. If you look in the left-hand side, there should be a new dedicated scheduled page in the ChatGPT sidebar that lets users view, pause, resume, edit, and delete all active tasks in one place. So like I said, with this, OpenAI is sunsetting Pulse, that personalized daily summary thing. And instead, everything is moving over to Tasks.
Jordan Wilson [00:04:31]:
So who has access? Well, this is starting to roll out now to essentially all paid plans. So before, to get this kind of power, you had to have that Pro plan for Pulse, but now it is rolling out to all paid plans. No word yet on when the free tier might get access, but there are rumors that this might happen. The cool thing is this is not just on the web, but it's also on iOS and Android as well.
So here's what OpenAI says about it. They said, we're making scheduled tasks easier to create and manage in ChatGPT with a dedicated scheduled page, improving task creation and editing flows, more flexible scheduling options, and better notifications. Users can schedule one-off and recurring tasks and also ask ChatGPT to check for changes and notify them when there's a meaningful update. So that part is new, as far as I know. You don't even have to tell it to check every single day at 9 AM. Let's just say that you're trying to monitor one of your competitors, as an example, because you know that they're going to be announcing a new product soon and there's a splash page and you want to know when it's updated. You can just say, check it out and tell me when it's updated. So that right there is a new piece.
Jordan Wilson [00:06:06]:
The cool thing, another thing that I like about this is they do have these kind of starter tasks that you can go in. A lot of times I think staring at a blank canvas or a blank prompt box is intimidating. That's why I liked the rollout of workspace agents a couple of months ago because it had a lot of good kind of templates. Same thing here with the scheduled tasks, and then you can also just kind of prompt them in the normal ChatGPT interface and then it will save there.
Another thing to keep in mind: this uses and can use all of your apps and connectors. So, simple things like triaging your email, calendar, and Google Drive every single morning. So if you have different files floating around a Drive, calendar appointments, all these things you need to do some research on, you can have this as a scheduled task every single day. If you've already connected, in this case, your Gmail, calendar, and Drive, you can have it run every single day, 6 AM, give you a rundown. Really cool. And then you can obviously play around with it in natural language to get it set up.
Jordan Wilson [00:07:08]:
So, I mean, why is it useful? Well, I think it's the combination of scheduling something, having this work proactively for you and using your connected apps. That's the big combination there. So I think for a lot of people that aren't Codex-pilled yet and you want this kind of taste of a proactive agent, this is enormous. It sounds small. It's like, OK, well, I could just go in and do this prompt when I need it. Yes, you can. But the biggest thing with AI now is you want it working for you. You should be prompting less and less in a prompt box. And you should instead be consuming more that agents are going out and doing on your behalf with your guidance.
So who's going to find this valuable? I mean, anyone. Whether marketing and ops professional, business operator, it doesn't matter. So one other thing to keep in mind: you can right now, I think, only create up to 10 tasks at a time. So we'll see if that changes. Hopefully it will. I would love to have hundreds. All right.
Jordan Wilson [00:08:18]:
Let's keep going. Our next one, Claude Design with some huge upgrades. All right. And let me say this. Claude Design, I think, was kind of polarizing when it was released. In theory, it was great, but also in practice, I think it failed in a lot of instances. We even did something, you know, designing PowerPoints. We did this a couple of weeks ago on one of our AI at Work on Wednesday shows, which will be back next week. FYI, we took a pause this week for our episode that happened to fall on Wednesday. But we showed you, once you know, designing PowerPoints in Claude Design, you know, it was fine, but a lot of the default everything in Claude Design looked the exact same, right?
So I think some of these new features that they announced will hopefully help in that and just make it a lot more useful.
Jordan Wilson [00:09:14]:
All right, so this was part of what Anthropic called the Claude Design June update. So here's what it is. This was just announced and it brings in a real WYSIWYG editing. That is the what-you-see-is-what-you-get editing, right? So if you've ever used something like WordPress or Wix or even Canva, that's technically a what-you-see-is-what-you-get editor where you can literally go click on something and edit it right and you get new options. I mean, that's what you have now in Claude Design, which is really, really cool and really powerful. This is one I'm stoked. This is one of the most excited releases that I've seen from Claude probably since artifacts. For me personally, I'm more excited to go use this new June update in Claude Design than I was with Fable. You know, Fable V was cool and all, but for a lot of my head-to-head use cases, it performed about the same level as GPT-4.5 Pro.
Jordan Wilson [00:10:22]:
This has that new, like I said, WYSIWYG editing, design system imports, two-way integration with Claude Code, which is great and I think was a sorely needed option from the original Claude Design, better export options, and desktop support.
So the design system you import into Claude Design is now the same component library that Claude Code uses to implement, creating a true round trip from prototype to production. And that's like what I said, that was one of those things that I thought was missing before in Claude Design. It just kind of lived there and it didn't even live inside of Claude Chat. It was just its own kind of interface. You had to click design and it opened up a new tab. But now it is more tightly integrated in Claude Code, which is awesome.
Jordan Wilson [00:11:08]:
This is, as an example, something I've been using a lot. People think, oh, inside ChatGPT or Codex, it doesn't have good front-end. And it's true, it doesn't. But if you use GPT-Image-2, which is the most powerful ImageGen model in the world by far, all of a sudden you do one prompt using ImageGen inside Codex and then just hand that off without doing anything, and all of a sudden you get great frontend. So now I think you have something similar here with the new Claude Design June update connected to Claude Code, which is really cool.
So who has access? So right now this is in beta on paid Claude plans, including Claude Pro, Max, Team, and Enterprise plans, included with your subscription. So it is still in that kind of sidebar or at claude.ai/design, and it shares usage limits with your other chat, co-work, Claude Code, all that.
Jordan Wilson [00:12:07]:
So here's why it's useful. Well, you can bring your own design system into Claude Design from a GitHub repo, design file, raw upload, and then Claude builds with your components and checks its output against the system rather than inventing its own buttons and spacing. So it also eliminates the rework loop between design and engineering. Prototypes can go directly to Claude Code without respeccing it, which is huge. And then some of the new export options, which I think are crazy powerful. So you can export to PDF and PowerPoint. That is the one small thing I said, OK, this is going to be big now. Because like I said, there were some things trying to design like decks inside Claude Design that just didn't work out. But now that you can export to PowerPoint, that's huge.
Or you can send work directly to third parties, like Adobe, Base, for Canva, Gamma, Lovable, Miro, Replite, Vercel, and Wix. So that right there, enormous, right? These tight integrations with all of these other kind of design or web platforms.
Jordan Wilson [00:13:17]:
For me personally, I'm going to be using the Canva one a lot. I'm going to at least try out the Gamma and Lovable one as well. All right. So here is what Claude says about the new updates. Claude Design now sticks to your design system across projects, works fluidly with Claude Code, lets you edit directly on the canvas, and connects to more tools you already use. It has a new home in the sidebar on the Claude desktop app. You can also find it at claude.ai/design.
So it says our new editor gives you direct fine-grain control over every element of your designs. New rich layout controls let you drag, resize, and align elements. Hundreds of stability fixes make the editor hold up under real use. So that's the thing that I think is going to be great. And then, like I said, all of those export options.
Jordan Wilson [00:14:38]:
So who's going to find this valuable? Well, anyone, right? So if you're in product design, development, marketing, right, anywhere. Also if you're using Canva, Gamma, and you're not really happy with the designs, the overall quality of Claude Design is stellar if you aren't using the defaults. Because like I said, the defaults are just, you know, so cookie cutter. So as long as you iterate a little bit in Claude Design, and now with the abilities that you have to better import your design system and export to other third parties, pretty big here. All right, let's keep moving.
Jordan Wilson [00:15:21]:
Next, we have a big tech player might be making a bigger splash in the AI avatars game. So we have new updates to Google Vids, and I think people are going to actually really like this. So here's what's new. Reading this from the Google Workspace update, they said with the integration of Gemini 3.1 Flash text-to-speech, which came out a couple of weeks ago, and the latest capabilities in Veo 3.1, AI avatars in Google Vids have become more realistic and expressive than ever.
They announced expanded language support, a new collection of avatar defaults, and the ability to direct your custom avatars to take action in any generated videos. So this is essentially combining a couple of more recent updates, specifically the Gemini 3.1 Flash text-to-speech, bringing that with Veo 3.1, Google's AI video model, into Google Vids.
Jordan Wilson [00:16:30]:
So a little bit more on what's new. That now has different avatar options. Instead of 23, there are 53 default presets spanning from photorealistic, 3D cartoon, and graphic novel styles. And video generations now run without duration limits, up from the previous 8-second cap. That's the thing that is going to make a big difference, I think.
So who has access? Anyone with a Google account in the US right now can try the new AI avatars in Google Vids at no additional cost. So the free tier limits to up to 10 Veo generations a month, and then higher quotas are available on paid workspace plans.
Jordan Wilson [00:17:13]:
So here's why it's useful. There's also an updated Google Slides integration that lets users convert presentations into more scalable, engaging videos with just a few clicks. That one is going to be big. Also, the custom avatars can now be directed to walk, talk, and interact with objects, simply by typing a text prompt describing their actions, making product demos and tutorials much cheaper to produce.
That's another big one because I think a lot of times these avatars, you know, they're kind of good, but they're kind of just clunky because they're just kind of robotic and not moving or sometimes you'll have these jerky hand movements. But now you can actually direct them inside Google Vids to pick up a coffee cup, take a sip, move your water from one side of the screen to the other, pet the cat on the desk, whatever it may be. So a lot of added direction there.
Jordan Wilson [00:18:15]:
All right, let's keep it going. This one from a name we don't mention a whole lot, but I think this is actually a pretty big update and you can go use it now. This is OpenRouter's new Fusion. OpenRouter Fusion lets you choose a panel of participant models alongside a judge model responsible for fusing the individual results together. With Fusion, running your prompt through multiple models in parallel, and then synthesizing a single output.
So right now the quality preset defaults, well, it did when it was released, to Fable 5 plus GPT-4.5. So whenever Fable 5 comes back, whether that's today, Monday, seems like it might be soon-ish according to the latest reporting, I'm guessing the quality preset will go back to Fable 5 plus GPT-4.5. And then the budget preset runs Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro.
So Fusion launched publicly as an experiment a couple of months ago, but now has been fully integrated into OpenRouter's API.
Jordan Wilson [00:19:30]:
So who has access? If you have an OpenRouter account, well, you have access. There's also a 128K context token window. So there is a way that you can go demo this and try it out. You can click Try Fusion Now. But I don't think that you can really do a lot of the more advanced features. You will have to have a paid plan for that, but you can at least go give it a shot, see if this works for you.
So here's why I think it's useful. You can synthesize the results of multiple models and that can significantly outperform what any individual model can produce on its own. So this is kind of like a mixture of models that I've been talking about for a very long time. We've seen offerings from Perplexity and Microsoft, like the Model Council. That's this idea, right? Having one prompt and then having it go through multiple models kind of back and forth and then having a judge look at it on the back end. This is something, if I'm being honest, that I do manually all the freaking time.
Jordan Wilson [00:20:54]:
So that's why this one with model fusion, I'm probably going to be using this a little bit just because it is so powerful to be able to use multiple models in different sequences or in different ways. A lot of times I'll have like an auditor, or using sub agents for this in Claude Code or Codex is really helpful as well, but this is just an easier way. If you're non-technical, you can go try it out at openrouter.ai/fusion. To run multiple models side by side, run an analysis and fuse into the best results.
Pretty cool here. You have a quality option, a budget option, and then a custom option as well. So who's going to find this valuable? Research and analysts running high-stakes, complex queries where output quality justifies the higher per-call cost. Developer teams that want to test cross-model consensus without building their own routing infrastructure. And organizations that need frontier-level reasoning without paying frontier-level prices on every single call.
Jordan Wilson [00:21:54]:
All right, let's go next. And we have artifacts again from Claude, this time inside Claude Code. So yes, Anthropic artifacts are not new. They've been around for almost 2 years now. And one of my favorite features that Anthropic introduced. And I think it was actually the artifacts feature that led to the very popular Gemini Canvas, which is still one of my most used features across any tool, and the ChatGPT Canvas as well, even though it's not called Canvas anymore, it's called Blocks. I miss Canvas. Bring it back, OpenAI. Just give me my toggle.
Anyways, artifacts actually set the stage for all of this, but it was really just available in the Claude.ai web interface. Not anymore.
Jordan Wilson [00:22:47]:
Let's talk about what's new in Claude Code artifacts. So Claude Code can now capture work progress as an artifact, turning Claude Code's work into live, shareable visual pages, including PR walkthroughs, system explainers, dashboards, and release checklists that update themselves as the session works. So Claude Code builds an artifact using the full context of your session, including the code base, connectors, and conversation itself, meaning a single incidents page can bring together failing tests, code monitoring, error spikes, and root cause reasoning without wiring up a separate data source.
So every artifact publishes to a persistent URL with version history and an org-level gallery for browsing and managing all artifacts created.
Jordan Wilson [00:23:37]:
So who has access to this? Well, you have to be on a Team or Enterprise plan using Claude Code, and it's accessible from the Claude Code CLI and the desktop app as well, with pages viewable in any browser. So yeah, right now, if you're on a normal paid plan or a free plan, you don't have access. It's only Team.
Also, this is kind of similar to Codex Sites, which is something I have been loving using. I do have a Claude Max. I don't have a Team plan, so this isn't one that I'm going to be using, at least not yet. I do think that this is a little bit more developer-focused than Codex Sites, which is a little bit more for or seemingly marketed toward your everyday knowledge work, where this artifacts feature for Claude Code seems to be a little more developer-heavy.
Jordan Wilson [00:24:42]:
So here's why it's useful. Well, it kind of replaces the walk-you-through-what-the-agent-found problem. Teammates can see the same live view with the same context in real time. And when Claude Code updates an artifact, the open pages refresh in place and teammates can then see the updates the moment they're published. So yeah, very similar to what OpenAI just released in Codex Sites about 2.5 weeks ago.
So who's going to find this valuable? I think ultimately engineering teams, right? If you are right now a heavy Claude Code shop, I think this is going to be great. Anyone that's on a Claude Team or Enterprise plan and doing a lot of software dev. Also product managers who want to track what actually shipped without asking engineers. I think that'll be helpful for them as well.
Jordan Wilson [00:25:36]:
All right. Two more quick ones. And the first was a sleeper but actually grabbing a lot of attention. So a new open weights model from ZAI. Yeah, actually, if you look at Artificial Analysis, it has a higher intelligence score than Gemini 3.1 Pro. This is the new GLM 5.2 from ZAI. So this is a 753 billion parameter mixture of experts model from the Chinese lab.
So it is text only. So this is not multimodal, and it has a 1 million token context window, up from 5.1's 200K, and it was released under an MIT license with weights on Hugging Face. So it is currently ranked as the leading open weights model on the Artificial Analysis intelligence index.
Jordan Wilson [00:26:37]:
So ZAI kind of positioned it specifically to dominate long-horizon autonomous coding and engineering tasks. And they said that it beats GPT-4.5 on multiple long-horizon coding benchmarks at roughly one sixth of the cost. So who has access to this right now? Well, anyone. So yeah, you can actually download it and start using it, but you're going to have to have like a supercomputer unless you want to run a very quantized version. But it's also available on their website at ZAI or the API if you're using it on the backend, which is much more affordable, as I said, than some of the proprietary closed options.
And it is fully commercial MIT license, which means it can be downloaded, self-hosted, fine-tuned, and deployed air-gapped.
Jordan Wilson [00:27:28]:
So for larger enterprises that are shifting away from token maxing to token efficiency, a model like this is huge. Also, it scored pretty off the charts on front-end design, front-end coding, actually even better than Fable on front-end coding, which is kind of bonkers. So yes, it is the best front-end coding model on the arena. Because Fable 5 is not technically available, GLM 5.2 is actually the best front-end coding model in the world and it's open weights.
So here's why it's useful and who will find it valuable. It's a frontier-class coding and reasoning model under a permissive MIT license that can run entirely inside your own boundary with no per-token meter and no dependency on a single vendor. The 1 million token context window, pretty big as well.
Jordan Wilson [00:29:02]:
All right. And about our last one here, which is the OpenAI Codex Record and Replay. All right, this one, we are back on track with OpenAI's kind of Codex Thursday releases. And this one is very simple but extremely powerful. So this was actually a feature that Google didn't roll out to everyone. They rolled it out to trusted testers. I think it was part of their Project Mercury, that just never hit the mainstream and didn't really hit at all. So OpenAI's Codex Record and Replay is exactly what it sounds like. It's a new feature in Codex where you click a button, you do your work, whatever a certain task or a series of tasks is, Codex watches, it learns exactly what you're doing, and then you can set that up as a skill or an automation that can run at will, any time.
Jordan Wilson [00:30:11]:
This is huge, right? A big step forward. Let's go ahead and read from what OpenAI said in their announcement post. They say, Record and Replay lets you demonstrate a workflow on your Mac and turn it into a reusable skill. Use it when the workflow is repetitive, depends on your preferences, or is easier to show than to describe in a prompt. For example, you might record how you file an expense, book a parking space, create a correctly configured issue, publish a video, or download a recurring report. Codex can package the pattern into a skill that you can use again with computer use, browser actions, connected plugins, or a combination of them.
So then it says, pick a workflow that you already know how to complete. Record and Replay works best when the steps are stable and the success criteria are clear.
Jordan Wilson [00:31:06]:
This one just came out hours ago. So I haven't had too much of a chance to play with it, but so far, really cool. And the Codex team did allude, actually in one of my comments on Twitter, that they might also, which would be great, bring the ability eventually to record your voice dictating what you're doing as well, which I think would be really helpful. So ultimately what this does is it creates a skill. So you can always go and modify it. If you go and run it one time and it doesn't work exactly how you want it, that's OK because you can still go in and update it.
So who has access? Well, it's right now only for Mac OS users and it's not available in every single country. So like a lot of these, if you're in the EU, the UK, Switzerland, you don't have access to this just yet. Otherwise it just requires a paid Codex plan, which if you have a paid ChatGPT plan, that's the same thing.
Jordan Wilson [00:32:22]:
So why is this useful? I mean, it removes the biggest friction in enterprise automation, having to describe a workflow in words when it's just much easier to do it. And then it converts that institutional knowledge, right? How, as an example, Sarah processes expense reports into a shareable, repeatable AI skill without IT involvement. So the cool thing is, I mean, this is already directly going to compete with RPA, robotics process automation tools like UiPath and Automation Anywhere on the recorded workflow and automated use case, but with a much lower setup bar. And it just works with all of your existing contexts inside of Codex, right? That's the biggest thing. If you already have existing projects, if you already have certain skills, certain workflows, your apps are all connected in there. This one for me, big.
Jordan Wilson [00:33:17]:
All right. So that is a wrap. Those are the 7 AI features that you should start using today. So as a quick recap: OpenAI, bringing back and unleashing more scheduled tasks to more users. The big Claude Design June update, which I am going to be using a lot. The new Google Vids AI Avatars update that's out now. OpenRouter's Model Fusion, pretty cool. Claude Code Artifacts, rolling out to Claude Code, not just on the web anymore. The new powerful open weights GLM 5.2 from ZAI. And last but definitely not least, OpenAI's Codex Record and Replay. I hope this was helpful. If so, please let me know by signing up at our website, youreverydayai.com.
Because like I said, you're missing out on a lot of these things if you're not constantly staying up to date. I do it for you. On Fridays, we do this Friday Features where we go over features that you can use right away. On Mondays, we bring you the AI news that matters. Those are big stories affecting your business, policy, the big AI LLM race. And everything on Wednesdays, we normally do demos going hands-on with one workflow, one new release from a company. And then Tuesdays, Thursdays, we switch it up. So thanks for tuning in. Hope to see you back on Monday and every day for more Everyday AI. Thanks, y'all.