Ep 808: OpenAI’s limited release of GPT-5.6, Mythos starts slow reinstatement, OpenAI gets spicy and more AI news

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life.

This June was supposed to be the start of hot AI summer.

That's because earlier this month, Anthropic released its Mythos-5 and Fable-5 models to the general public, and we were supposed to have an avalanche of other front tier models released to all. Google was slated to release Gemini 3.5 Pro, and OpenAI was on board to drop GPT-5.6 all this month. And although we technically did kind of get the GPT-5 release from OpenAI late Friday, it's not exactly the hot AI summer we were promised. That's because Anthropic's heavy capabilities hypings of Mythos caused the US government to step in, and now we're at the end of of June with no general availability to the promised frontier models from any of the big three companies. Although we do have some updated news because Mythos 5 and the new GPT-5.6 have started to roll out, but only if you work at a select few companies. So that's the big AI news of the week, but there's a lot more that we're going to get you caught up on. So if you don't have hours every single day trying to understand what's happening in the AI world and how impacts your business, don't worry. That's what I do every single day. And then on Monday, we bring you the AI news that matters. So let's get into it. If you're new here, welcome. My name is Jordan Wilson and this is Everyday AI. It's a daily livestream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up, but how we can use all of this information hidden between the headlines to get ahead and grow our companies and careers. So it starts here. With the unedited, unscripted livestream podcast. But please make sure if you haven't already, subscribe to the podcast on Apple Podcasts and Spotify, and then go to youreverydayai.com. We're going to be recapping all of the highlights from today's show as well as all of the other AI developments you need to know to make good decisions for your company. And yeah, there's going to be a lot new, I'm guessing, updates from today, uh, with the updates to the Anthropic Mythos release, then pullback, then re-release. And now we have GPT-5.6 and 3 new variations. So, uh, let's start there, shall we? Uh, because we do have a new leading model from OpenAI, but unfortunately looks like hardly anyone has access to it. Uh, so OpenAI says it is limiting its early access to its newest GPT-5.6 models to a small group of trusted testers after requests from the US government, raising new questions about how much control Washington should have over advanced AI releases. So OpenAI says the first preview of GPT-5.6 is restricted to partners whose participation has been shared with the US government rather than their normal kind of release cycle of just opening it up to millions of users right away. So aside from this kind of new-ish way to roll out models, which is just going to selected few, OpenAI also has a new naming scheme. So here we go. We have SOL, uh, which OpenAI calls its flagship and most powerful model. Then we have TERRA, a more balanced version for everyday use. And then we have LUNA, which is a faster, lower-cost option. So, uh, similar to the older Anthropic 3-tier naming capabilities, it looks like that's what we're going to have from OpenAI moving on. So, uh, What that means is, well, you have your Sun, your Earth, and your Moon in that order. So SOL is the, uh, the big heavy, uh, model. Uh, then you have Terra, which is kind of the in-between or the old SONNET from Anthropic if you're used to that naming mechanism. And then you have Luna, right?

Our Moon.

Uh, that's the faster, lower cost option. So yeah, think of your globe, uh, you know, Sun, Earth, Moon. Yeah, we're gonna have to start memorizing these things. So OpenAI says also that SOL adds a max reasoning mode and an ultra mode that uses coordinated subagents for very complex tasks, especially in coding, biology, and cybersecurity. So the company says GPT-5.6 SOL is slightly better at coding workflows than Anthropic's Claude Mythos-5 while using about one-third of the output tokens. That is the big news there. So, uh, we don't have full benchmarks on the new GPT-5.6 SOL, but it looks like from what OpenAI shared, it does look like it's in the Mythos and Fable range on some benchmarks. On some benchmarks, Anthropic is ahead. So we'll see when it comes to third-party analysis like artificial analysis or the LM or formerly LM arena. So we'll see where it clocks in, but it looks like it's probably a little bit below kind of true Mythos level. So maybe it's more in the Fable 5 area, which is Mythos without all— or with— that is Mythos with guardrails. So yeah, it's super confusing now, right? Because 3 weeks ago we didn't have Mythos and Fable and SOL and Terra and Luna, right? So now we have not only this new tier of maybe AI not being truly democratized anymore, and it's going to be You know, who knows if the general public will ever get access to Mythos. Maybe it'll just be a limited availability with Fable. And seems like there's some rumors circulating that, you know, in the future you might actually have to show US citizenship to use some of these models. So we'll see where the US government lands with that. But it is going to be one of the hottest topics. But regardless, it looks like a very promising release from OpenAI. So the Trump administration has tightened oversight of Frontier AI releases, and Dean Ball, a former White House AI advisor and now recently hired by OpenAI, says a recent executive order has effectively created an involuntary licensing system for most advanced models. So Ball argues that if the government does not define clear safety guardrails, AI launches could face repeated delays, which could slow US competition with China and complicate larger infrastructure investments. So OpenAI says its new models include stronger built-in security with defenses meant to resist jailbreaks and to prioritize defensive cybersecurity to help offensive misuse. OpenAI also says it is moving away from a separate filter system and toward safety guardrails built in the model itself after backlash over Anthropic's earlier approach of routing risky prompts to older models. So Yeah, we'll see, right? OpenAI said they hope to roll out access to the new GPT-5.6 family to everyone in the coming weeks. We'll see where we land with that because Anthropic, as an example, when they had to pull Fable 5 access, which they released to all paid users, they said that they would have some sort of communication update in 24 hours. It's been more than 2 weeks and we don't have an official word from Anthropic. So OpenAI kind of promised promised seeing the GPT-5.6 rollout to more people, uh, in the coming weeks. They were pretty, um, pretty forthright in their announcement talking about some of these complications, uh, of rolling this model out in kind of this new reality we live in now, apparently, which is, uh, AI is no longer, at least temporarily, maybe permanently, truly democratized. And there may be a tiered system in terms of who has access to what level of intelligence. All right, sticking with a similar story. Well, Mythos is starting to be re-rolled out, but only to trusted users. So, uh, yeah, a lot of action on Friday. So this was, I believe, shortly after, uh, OpenAI announced GPT-5.6 and its limited release to trusted testers. So now the US government on Friday, uh, apparently is allowing Anthropic to re-release its Mythos 5 models to only about 100 companies and a federal agency, marking a key shift after a week standoff with the Trump administration. So the Commerce Department, according to reports, said in a letter that appropriate safeguards are in place for certain trusted partners to access the Claude Mythos-5 models, which suggests the government is still limiting broad access to Mythos-5 and, well, Fable-5, because the approval does not cover Anthropic's Fable-5 model, meaning the company still does not have permission to restore access to its newest, most advanced release. Obviously that could change any hour, any minute. So, you know, make sure that you check today's newsletter. It's funny because actually on Friday, you know, we always do A/B tests in the newsletter. So I sent the A/B test and then literally 3 minutes later we got the release or the news from OpenAI. So, you know, we'll see if there is any new update, at least when it comes to Anthropic. But Anthropic has disabled access to both Fable 5 and Mythos 5. Earlier this month after an export control directive cited national security concerns and ordered access suspended for foreign nationals, which included foreign national employees. So the move came only days after Anthropic unveiled the models and said Fable 5 was its first public release of such an advanced system backed by new safeguards the company said blocked responses in high-risk areas. So essentially, Anthropic, after that directive from the US government, said, well, there's no way to actually do that. So they had to pull the model worldwide. And that's where we still stand. Aside from this new update of apparently the US government is letting those roughly 100 companies and federal agencies get access back to Mythos. So Anthropic's negotiations with the Trump administration have also reportedly shifted from CEO Dario Amati to co-founder Tom Brown. Brown, underscoring how serious and sensitive the talks have become. So, yeah, this one was kind of interesting to see. According to reports, there were some White House officials that alluded that Dario was being a weirdo and they couldn't really get talks going with him. So instead, Anthropic co-founder Tom Brown was tapped in for negotiations. And it seems like, at least from Anthropic's point of view, things are going better than they were at least last week or for the rest of June. Uh, the company has also faced pressure from the Defense Department, which previously labeled Anthropic a supply chain risk, a rare designation usually reserved for foreign adversaries. Anthropic has sued the Trump administration over that original, uh, blacklisting, and the legal fight is still ongoing, adding another layer of uncertainty around the company's government business.

AI moves too fast to follow, but you're expected to keep up. Otherwise, your career or company might lag behind while AI-native competitors leap ahead. But you don't have 10 hours a day to understand it all. That's what I do for you. But after 700+ episodes of Everyday AI, the most common questions I get is, Where do I start? That's why we created the Start Here series, an ongoing podcast series of more than a dozen episodes you can listen to in order. It covers the AI basics for beginners and sharpens the skills of AI champions pushing their companies forward. In the ongoing series, we explain complex trends in simple language that you can turn into action. There's 3 ways to jump in. Number 1, go scroll back to the first one in episode 691. Number 2, tap the link in your show notes at any time for the Start Here series, or you can just go to starthereseries.com, which also gives you free access to our inner circle community where you can connect with other business leaders doing the same. The Start Here series will slow down the pace of AI so you can get ahead.

All right, so back to the other AI company as part of the Big Four, right? We have OpenAI, Anthropic, Google, and Microsoft. So we just talked about where OpenAI and Anthropic stand with their latest models. Now, what about Google? Because at last month's Google I/O conference, when they announced and released their Gemini 3.5 Flash, they said that their big model, Gemini 3.5 Pro, would be unveiled in June. And here we are in the last hours of June, and we don't have any new model model from Google. Yep. So according to reports, it is kind of the pressure now from the US government to get this model approved. But there's also allegedly some other things going on behind the curtains at Google. So according to reporting from The Information and other outlets, Google is turning its short-term AI coding strike team into a permanent mid-training group to help close the gap with Anthropic in AI-assisted coding development. So, uh, this move matters because Google appears to be reshaping how it trains Gemini, shifting from relying mostly on broad pre-training and fine-tuning to adding a formal middle training stage with specialized coding data. So that stage is called mid-training, and it's meant to teach the model more than code syntax, including how to work through larger engineering tasks, tool use, and multi-file code understanding. So the reorganization follows pressure from Anthropic's Claude Code, which has become a major competitor in AI coding, where internal estimates reportedly put Google engineers at about 50% AI assistance versus nearly 100%, reportedly at Anthropic. The timing is also tied to key talent loss, with several notable researchers in engineering and engineers leaving Google for rivals, including, uh, Jonas Adler, uh, Alexander Pritzl, John Jumper, and Noam Shazir, according to reporting. So yeah, Google has kind of lost an all-star team of AI engineers over the past few weeks, uh, which led to them reportedly turning this strike team, right? So it's kind of a different variation of these code reds you hear these companies, you know, push out internally when it's like, oh, it seems like we're kind of behind, we better call a code red. So that's kind of the equivalent of what this strike team is, right? They said, hey, a couple months according to reports, that they realized internally that they were falling behind on coding. So they developed what was supposed to be a temporary coding strike team. And now apparently, uh, that, uh, you know, kind of realization that maybe they're a little further behind on coding— this is according to reports, not according to me or my views, uh, right? So that combined with losing some major Talent. I mean, those are some names if you follow AI. I mean, those are some of the OGs of AI. So Google originally formed that team in April to speed up Gemini's coding abilities. So yeah, it could be a long-term thing. And the updated roadmap now goes beyond simple code completion and is aimed at long-running engineering agents that can carry out multi-step tasks over time with less human supervision. So yeah, we'll see, uh, what happens, and we'll be continuing to follow, uh, all these developments from Google because, yeah, they did publicly say in May that Gemini 3.5 Pro would be out next month, which is June, right? And how many days we have left of June? We have, uh, I don't know, about 30-some hours left of June. So we'll see if we get an official announcement from Google. I would expect No, we won't, just because of this new climate that we are now living in. All right. Our next piece of AI news. A group of AI juggernauts and former government workers have combined to create a new nonprofit helped at, well, raising up the employability of workers. So a new nonprofit called Raise US says it has already raised $500 million toward a $1 $1 billion goal to help states in the US prepare for AI-related disruption. So the group is being led by former Commerce Secretary Gina, Gina Raimondo, who said the US has a tech strategy for AI competition but still lacks a clear people strategy for workers. So RAISE-US is backed by major players in tech and AI, including the OpenAI Foundation, Anthropic, Amazon, and Microsoft, which are serving as anchor partners. So the nonprofit says it will build a national platform to advise state governors on how to prepare local workforces for AI-driven change, with a focus on practical policy and training models. So far, there are some notable state partnerships already, right? Kind of hours after we got wind of this new group. So the first state partnerships are all already underway with Arkansas, Connecticut, Maryland, and Utah split evenly between Republican and Democratic-led states. In Arkansas, Arkansas, Raise.US is working with Governor, uh, Governor Sarah Huckabee Sanders on Arkansas Launch, an AI-powered career navigation platform that would connect students and job seekers to personalized learning and employer-linked career paths. In Maryland, the group is working with Governor Wes Moore to expand its service year opportunities for recent high school graduates in fields like healthcare and education. Utah is a notable test because the state has has been forced into the national spotlight over the buildout of AI data centers, showing how AI expansion can create both job and infrastructure tensions. So RAISE US says more states are expected to join in the coming months and also wants to test policies such as short-time compensation and wage insurance that could soften the blow of job changes. So the effort comes as debate continues over how much AI will affect the labor market. As we know, we heard, uh, kind of some, uh, wild, uh, kind of predictions such as that from Anthropic CEO Dario Mati, who originally said that AI would disrupt up to 50% of white-collar jobs, which obviously didn't happen, and he kind of had to adjust or update his stance on that. But regardless, we are already seeing in 2026 an unprecedented, uh, impact of AI on full-time jobs. So, uh, it's good to see a company come in and address the human side, which I've always said since I've started this podcast 3 and a half years ago, that AI implementation is much more about people and man— uh, people management, change management than it is about, you know, traditional tech implementation or digital transformation. So, all right, our next piece of AI news, we have a fresh accusation of Anthropic pointing their fingers at the Chinese saying, you stole our model.. So according to CNBC, Anthropic sent a letter this month to the Senate Banking Committee accusing Alibaba of carrying out what it called the largest known distillation attack on Anthropic's AI models. So Anthropic said the activity involved about 28 million exchanges through roughly 25,000 fraudulent accounts between April and June, which it says points to a large-scale attempt to copy its model behavior. So the company told lawmakers that operators linked to Alibaba and its AI lab were behind the effort, framing the issue as a major example of AI model theft rather than normal competition. So distillation is a process, if you don't know, where a smaller model learns from the outputs of a stronger one. But Anthropic is accusing the Alibaba-linked effort of using that method to, well, steal all of their work and infuse it into Alibaba's models. Anthropic said fighting this kind of AI copying will require coordination between government and industry, and that it wants Washington to help protect US leadership. So, uh, according to reports, the letter comes 2 months after the White House Office of Science and Technology Policy said it would help AI companies in the US detect and respond to industrial-scale distillation attempts. Anthropic topic said Alibaba's actions ignored the Trump administration's warnings, uh, making the disputes part of the bigger US-China technology and national security fight. So this is definitely a big story to keep a watch on because, well, this also impacts what models your company may or may not be using. So last week, make sure you go check out, uh, our Start Here series if you haven't already.. But in episode, uh, which one was that? In episode 29 of the Start Here series, which was episode 804 of Everyday AI, we talked about a new open source model called GLM-5-2, which may be the first time we have a Chinese open weight or open source model that is almost at the frontier level. So it's going to be pretty interesting. And I think this, uh, probably deserves a larger conversation. If the access to true frontier models is no longer actually democratized, right? If 99.9% of US workers, at least right now, right, can't access Anthropic's Fable, they can't access, uh, Anthropic's Mythos, they can't access right now, at least, maybe that might change in a week or two, they can't access, you know, GPT-5.6, So when you have these other Chinese models that are starting to get in that tier, right? I wouldn't say they're in the tier of GPT-5.6 and Mythos, but they're in the tier right below, right? Which is, well, models available to every single person, right? So that's GPT-5.5 and Opus 4.8 from Anthropic. So as you get these Chinese models, right? Whether they're from Alibaba, uh, they're from ZAI, whoever they are that are starting to get in that tier of the best publicly available models. It is an important conversation to keep an eye on. So distillation, it's, it, the combination of distillation and maybe these frontier systems only be being available to the select few, creates for a very interesting and mixy, sorry, a potentially explosive combination, right? In terms of what models your company should use and well, what it means for revenue of companies like Anthropic who are preparing to go public. All right, our last big piece of AI news this week is chip news. Yeah, because OpenAI and Broadcom have unveiled Jalapeño, a custom chip meant to run large language models more efficiently, with deployment expected to happen later this year. So the big picture goal for OpenAI and Broadcom was to make an AI systems, uh, to make their AI systems cheaper and faster and less power hungry to run at scale, which matters a lot because energy and computing costs are becoming major limits for AI companies and their ability to provide frontier AI models to the broader enterprise lens. So with this new chip partnership, OpenAI's first, OpenAI says Jalapeño is built specifically for AI inference, meaning it is designed to help models answer questions and serve users rather than train them, right? So obviously on the training side, NVIDIA's GPUs have been the industry standard for a long time. This chip is specifically built for inference, like all of the back and forth that OpenAI needs when, uh, up to a billion users each week are using their services. So the companies say the chip is aimed at the hardest parts of the AI service, including speed, speed, memory use, networking, and reducing wasted compute. So OpenAI and Broadcom claim the chip will be more efficient than current leading hardware, but they have not shared benchmarks or hard performance numbers yet. Images shared by the company suggest a very large chip with multiple memory modules, which points to a serious effort to handle heavy AI workloads. So the companies say early samples are are already working in the lab, which is a crazy fast turnaround. And OpenAI says it is testing real workloads on the chips now. OpenAI and Broadcom also say the chip was developed in just 9 months, which is much faster than typical, uh, typical chip development and shows how quickly AI infrastructure is being pushed forward. All right, so that's a wrap for our major big stories of the week. But let's quickly move into what's new and what's next. So this is a bullet point roundup of everything else noteworthy that either happened last week or is happening this week. So it's a mixture of smaller news stories, maybe some leaks and rumors, but let's get straight into it fast, bullet point style. Here we go. So nearly 400 newspapers have sued OpenAI and Microsoft over content. Qualcomm and Meta announced a multi-generation data center, uh, CPU partnership. Uh, Chevron signed a 20-year Microsoft Power deal for AI data centers. Uh, Meta launched a $299 pair of AI glasses, uh, in line with SLR Luxottica, the famous eye brand maker that I probably did not pronounce correctly. Uh, Google reportedly limited Meta's access to Gemini AI models over capacity shortages. Elon Musk Fox says that XAI's new 1.5 trillion parameter Groq 4.5 has begun private beta testing. So we'll see if they have to go through the normal, uh, government, uh, hoops, uh, jumping through those, or if they'll be able to release it. Uh, Micron and Anthropic sign a memory deal to help scale Claude. Apple's VP for Vision Pro and smart glasses is leaving for OpenAI AI's hardware unit, according to reports. Uh, leaks show that OpenAI is testing computer use with Excel and PowerPoint add-ins in Codex, which will be an interesting update. Uh, Google inside of Notebook LM added fully customizable flashcards for users to edit and share. OpenAI reportedly is weighing a 2027 IPO instead of 2026 after some recent SpaceX volatility. In an internal study, OpenAI said Codex accounted for 99.8% of internal weekly AI output tokens, which is a wild percentage. Almost 100% of their output tokens internally are in Codex, not in their ChatGPT service. OpenAI released Daybreak Cyber tools, including GPT-5.5 Cyber and Codex security updates. OpenAI released a refreshed GPT-5.5 Instant model, with better conversation and handling in constraints. Google launched Google Finance in general availability with portfolio tracking and AI research tools. Google released Gemini for Home smart speakers for natural voice conversations. Anthropic launched Claude Tag in Slack, which is kind of like a manager per Slack channel that can do anything that Claude can do. Canva released Canva Grow 2.0 for AI ad creation, publishing, and optimization. Google added Select from Screen in Chrome's Gemini sidebar after they updated their Gemini 3.5, uh, Flash to have computer use capabilities. Microsoft added Copilot Skills in Excel for repeatable task automation, and a new DigitalOcean plugin for Codex entered public preview for remote cloud dev machines. All right, that's a lot, and I can almost guarantee you this week is going to move even faster than last week. Because y'all, this everything with OpenAI and Anthropic in terms of GPT-5.6 and the three-tiered system, the government kind of asking them to hold it back, the re-release, the slow re-release of Anthropic Mythos, all of this just happened Friday. So I expect a ton of new developments that you need to be paying attention to. So don't spend hours every single day sitting around your boardroom talking about AI strategy and looking at that day's news. News, right? Just let me do that for you. Then join us on Mondays as we bring you the AI news that matters. If you're new here, on Wednesdays we go hands-on. We usually do a demo. So make sure you check out today's newsletter so you can let us know what you want to see on that show. On Fridays, we go over fresh AI features that are available to everyone. And then on Tuesdays and Thursdays, we rotate the type of show. So please, if you haven't already, subscribe to the show on Spotify and Apple Podcasts, then go to youreverydayai.com. Sign up for the free daily newsletter. Thanks for tuning in. We hope to see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.