302 | ChatGPT 1B user milestone but slips below 50% market share, GLM 5.2 takes the lead in coding, Microsoft CoPilot Cowork now available, and more AI news for the week ending on June 19 2026

0:00Hello, and welcome to a weekend news episode of the Leveraging AI podcast, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business, grow your business, and advance your career. This is Isar, Metis, your host, and we had a very interesting week this week. Nothing crazy major happened, but lots and lots and lots of small things did happen, and we are going to talk about all of them in this show. So this is going to be more of a rapid-fire, beginning-to-end kind of session, but in bulks of specific topics. The first topic we're going to talk about is the burning competition in the AI space and a few interesting things that happened this week.

0:38We're going to talk about AI security and governance as a big topic as well. As it is becoming a bigger and bigger issue, we are going to talk about agentic platforms. Again, not surprising. I think that's going to be a weekly thing from now till forever, or at least in the foreseeable future. So we have a lot of things to cover. So let's get started. The first topic I want to talk about is that OpenAI just reached a huge milestone, which is reaching over a billion monthly active users in just three years after it has launched its platform, or three and a half years if you want to be more specific, which is significantly faster than any other platform in history.

1:21So Facebook took eight years to get to a billion users. YouTube took eight years to get to a billion users. TikTok took five years to get to a billion users. And Chachupiti did this in about three and a half years. Now, that being said, Sam Altman has been very clear in the past few months that profitability doesn't seem something is even on the horizon. But with that being said, it is very unclear how OpenAI is actually going to become profitable. And if you want, the quote from a few weeks ago from Sam Altman, who said, and I'm quoting, no current plans to make revenue and no idea how we may one day generate revenue.

2:00And his plan is basically once they reach AGI, asking the AGI to give them ideas how to become profitable. But what he's saying, that's not his worry. His worry is about developing better and better AI solutions to support humanity. And so when you hear this, when a company is growing to a billion users and still bleeding crazy amounts of money on the year they're on the verge of IPO, it is very interesting to see how the IPO is going to go forward. we're going to talk a lot more about this as we hear multiple sides of the OpenAI story. To me, the bigger story that nobody talked about when they were mentioning the 1 billion users, which is definitely impressive, is that they reached 900 million users and crossed that sometime in beginning of Q4 of last year.

2:46So they were growing significantly faster before, and they kind of plateaued since. And while 1 billion users is a huge milestone, it seems to be on the slowdown of adoption of ChatGPT specifically, definitely not AI as a whole. Now, to make it more specific, another big milestone happened this week, which is ChatGPT's global market share decreased below 50 % for the first time. So in the beginning of May, which was the last time we got this number, they were just above 50%. And now the latest news at the end of May is 46.4 % global market in the AI space. That is still a huge lead over number two, which is Gemini at 27.7, but it is definitely a shrinking gap from a percentage perspective.

3:37And if you're wondering about where Anthropic is in that mix, there are 10.3, seeing an explosive growth, but still from a market share perspective, significantly smaller than Chachupiti and Gemini. So Claude's year-over-year growth has been 640%, while Chachipiti's has been 62%. So 10x faster growth in percentages. Again, in total numbers, they are still significantly smaller, or not in total numbers, but in total percentage of global market share. All this data is basically based on Sensor Tower report that just came out. now if you're wondering where are all the other players such as grok from spacex previously xai and perplexity and meta and deep seek and all of those collectively account for less than five percent of the total ai assistant market space right now which is basically telling you it's a three horse race and that's it it is chatupiti gemini and claude and it is very obvious that in this race right now, Chachupiti and Claude from Anthropic are focusing completely on enterprise and businesses, while Gemini is still more focused on consumer and they're slowly making moves into the enterprise space.

4:51It wasn't like a code red like happened in Chachupiti a few months ago at the beginning of the year where they flipped everything they did and then they canceled Sora and took very clear actions to push codecs and focus on enterprise and code usage. By the way, the Tuesday episode this week is going to be how to use codex for things that are not creating code, which I think you will find a lot of value in if you did not jump ship yet to the Cloud universe and you're still in ChatGPT and just want to grow in the agentic direction. Or if you have moved to the Cloud universe and you're considering what's happening on the OpenAI side, it will definitely open your eyes on how much progress they made with codex.

5:29But back to the news, another thing that OpenAI is doing right now to aggressively push, their dominance or their market share in the enterprise space. They just announced that they're launching a partner network and they're investing$150 million in that. So on June 14th, they announced OpenAI Partner Network and that they're going to be investing, like I said, hundreds of millions of dollars to help enterprises move beyond AI pilots into real world impact. In fact, they are in the process of training 300 ,000 consultants by the end of 2026. That's a very aggressive timeline. The network launch partners are the elite of the elite.

6:09It's Accenture, Bain, McKinsey, PwC, all the obvious suspects. And the goal is to create a huge capacity to help and implement AI across different companies around the world. The partners progress through select, advanced, and elite tiers based on their sales performance, technical capability, deployment experience, and with specific specializations in different areas such as codecs, cybersecurity, AI agents, etc. So it is an entire training program that is designated to help other companies, again, consulting companies, help other businesses implement specifically OpenAI's platforms in a successful way.

6:48They've shared multiple success stories with, for me, the most exciting and impressive one is in early deployments, Paycheck achieved 80 % reduction in wait time and 30 % reduction in effort for payroll processing. that is very significant. And I can tell you from the work that I am doing with companies, so I'm working with multiple companies around the world from small businesses to large enterprises and holding companies and helping them implement AI through multiple platforms, not specifically open AI. And the benefits that I'm seeing are incredible. And the biggest benefits come when you train the actual people in the organization on how to implement AI themselves versus hiring consulting companies to do it for you.

7:33Now, I'm not saying there's anything wrong with hiring consulting companies, but the benefit of having employees from multiple departments know how to do this themselves is a huge benefit because they themselves are the people doing the work. They understand the pain points, they understand limitations, they understand the exact processes, and their ability to create magic is absolutely incredible. And again, I'm seeing this time and time again. I just did a workshop this past week to a large organization. This was the second workshop in two weeks for the same organization. And between these two workshops, one of the people came and developed an incredible dashboard that takes together information from multiple sources and is going to dramatically improve their efficiency and reduce the amount of fees they're currently having and improve their cash flow by being able to deliver what they're delivering in a more effective way, taking into consideration multiple priorities and data points.

8:23He did it on his own and he did it a week after the first training that he got before the second training. And the reason he was able to do this is the fact that he understands exactly what the pain points are. He understands the data and he understands what benefits he can gain by doing very specific things and getting into specific insights, which will be very hard for a consultant to understand. I mean, it's doable, but the consultant will have to interview, review, and create programs and plans and do a lot of things that are just overhead when the people are doing the actual work, just understand what's going on.

8:57And they can do this very effectively if they are taught how to do this, which is exactly what I'm doing. And so I highly recommend to you, if you are looking for ways to accelerate and improve the efficiency of your organization, don't mess up on this really incredible technology. and come and find somebody, it doesn't have to be me, but to train your actual people on how to implement AI effectively because the results are life-changing. It is really very, very different than anything we've seen so far. But now back to the news and back to OpenAI and why they're doing what they're doing, which relates back to something you heard me say multiple times in this show, is that OpenAI themselves has stated that the model capabilities are no longer the limiting factor.

9:37I've been saying this for at least a year now. All the models are good enough to do probably 80 or 90 % of the knowledge work we're doing today, if you know what you're doing, if you can connect your data effectively and securely, etc. And so this is exactly why they've established this group or this partnership program, so they can have other people help companies solve that, including change management at scale, including driving adoption, including all the things that are not related to how powerful or capable the model is. Staying on the topic of adoption and competition in this space, Microsoft and Anthropic finally made Microsoft Copilot Cowork generally available.

10:16So following the initial preview launch in March of 2026, during this preview period, just a handful of companies, when I say handful, it's over half of the Fortune 500 companies, but we, the common people, did not get access to that, got access to Copilot Cowork, which making it per Microsoft, the fastest feature in Microsoft Frontier program ever. And they're claiming that they're seeing significantly high user satisfaction from this platform. Now, those of you who don't know what Copilot Cowork is, it's basically taking Anthropics Cowork, which is the platform that I've been using most and the platform that I've been teaching how to use and connecting it inside the Microsoft 365 universe, which makes it extremely powerful.

11:01So all the things that you can do in the regular co-work, you can do right now inside the Microsoft environment, enjoying the security and data access that the Microsoft environment provides. It connects together Outlook and Teams, Word, Excel, SharePoint, and the agentic capability connects to Microsoft WorkIQ context engine, which allows you to understand what's happening in the organization from multiple data sources. So while I think Cloud Cowork is an incredible platform, literally the most powerful tool I've ever used. Microsoft Copilot Cowork should be even more powerful if you are in the Microsoft environment.

11:38And as I mentioned, it is now generally available. So the courses that I teach and the workshops that I do for companies that focus specifically on that will allow you to do incredible things if you are not in the Microsoft world, but be even more powerful if you are in the Microsoft universe, which is very exciting for Microsoft users. It is very exciting for Anthropic. And it is very exciting for me because it is a really, truly amazing process to see the transformation in companies that learn how to use this effectively. Now, in addition to the release, they introduced new partner plugins with nine immediately available, including Monday.com, Miro, Moody's, NO6, Harvey, LS Morningstar, S &P Global Energy, and Teams Maestro.

12:21And there are more coming from Adobe, Atlassian, Box, Canva, and more. So in addition to being able to connect to the Microsoft environment, you'll be able to connect to more or less everything in your tech stack, either already or in the immediate future. Now, this new tool, as expected and sadly, comes with usage-based billing versus just the license billing. So you still need to have a subscription license to Microsoft Copilot 365, but you also are going to pay per usage for using Copilot. This is similar in all the enterprise platforms that are out there, which adds up relatively quickly, which means, again, if you learn how to use this effectively and use tokens wisely, you will be able to get better results while spending less money.

13:06Staying on the changes of the competitive landscape, SpaceX just agreed to acquire the AI coding platform Cursor for$60 billion in a stock deal, just a few days after SpaceX's historic largest ever IPO. Now, this deal is not new. It was on the table for a while. There were discussions and talks about different numbers for a few months now. So it's not new news. But the fact that it is now happening is definitely news. The acquisition cost, as I mentioned, is going to be$60 billion. SpaceX IPO stock price was$135. They just crossed$200 in pre-market trading, adding nearly$1 trillion to the valuation since the stock became public, and it is just 16 times cursor evaluation just in the past week and a half.

13:59This comes at the tail of very significant changes that were happening in XAI that then became a part of SpaceX. So all 11 XAI co-founders, other than Elon Musk, departed by the end of March of this year, with Musk himself admitting that XAI was, and I'm quoting, was not built right first time around and requires rebuilding from the foundations up. One of the main things that Elon was trying to push in the past few months is better coding capabilities inside of X. And now he's going to get it because he now has Cursor, all of its IP capabilities and engineers as part of his companies. Now, I mentioned to you before, I think SpaceX will probably play a much bigger role in the hardware side of the AI race versus the software side of the AI race.

14:46But this move and investing 60 billion, which again is a drop in the ocean compared to the current valuation of SpaceX, especially that this is a stock only deal, meaning they're not exchanging any cash, is going to allow SpaceX to be significantly more active and in the software side of the AI race. Now I want to talk about additional aspects in the race, but it is actually the impact of the US government stopping Claude Fable and Mythos 5 from being released to the world, or actually not stopping it from being released, but it was released and then pulled back three days later. And what happened in the following week and a half is we got a handful of extremely powerful open source models that are becoming a real option for companies who want to adapt advanced AI.

15:37So within a few days from the pullback of the latest models from Anthropic, we got new models from Cohere, Moonshot, and Zifu all back to back. So a quick reminder, on June 12, 2026, the US government via export control directive has ordered Anthropic to suspend Fable 5 and Mythos 5 for all foreign nationals. Since this is very, very hard to do, they just pulled it from everybody. So we all had, or if you're an Anthropic user like me, we had access to Fable for a few days and then it will pull back. And now we're back with 4.8. And that's despite the fact that Anthropic has disputed the severity of what the government has called as a big issue.

16:16Anthropic said it's a narrow, non-universal jailbreak and that it is very similar to the capabilities in models like OpenAI GPT 5.5. That didn't help them yet, even though in the latest answers to reporters in South Korea, Anthropics' Chris Cloris said that he believes or he's confident that the model will return in the next coming days. But back to the open source models, between June 9th and June 13th, several open-weight coding models rapidly emerged as direct alternatives. So Cohere launched North Mini Code, which is under an Apache 2 license. Moonshot released Kimi 2.7 Code, and Zifu introduced GLM 5.2.

16:56Now, before we dive into all of them, the most impressive one is GLM 5.2, which now ranks number two on Arena.ai Code Arena front-end leaderboard. So this is, we talked about this many, many times. It's a leaderboard that is based on actual people using the platform doing a blank test. So you give it a prompt, you get two results and you pick the one that was more helpful to you. And they're now ranked second only to the now not available Fable model from Claude. And then it is actually better than Opus 4.7, 4.8 and GPT 5.5 on that platform. So this is an open source Chinese model that on its max settings is outperforming every available Western closed source model in code generation.

17:44This is very, very significant. So a little more about this model. It has a 1 million token context window that they're saying is not just a number on paper, but actually battle tested in actually using this context window for large scale code operations. This is a 5x increase from GLM 5.1, which had a 200 ,000 context window limit. This is, again, a very significant from the capabilities it provides. It is doing extremely well on all the coding benchmarks. It is only four points behind Cloud Opus 4.8 and outperforming GPT 5.5, as an example, on Frontier SWE, which is one of the key benchmarks that platforms are being tested on when it comes to writing code.

18:28It comes with flexible level of thinking, so you can adjust or it can adjust on its own the effort levels balancing between the performance you need and computational costs and it is significantly cheaper than its western competitors. In addition, Moonshot AI launched Kimi 2.7 which is an extremely powerful open source coding focus agent model designed for long-term horizon software engineering. Another tool to do the same thing. It is reducing the thinking tokens compared to their previous model by approximately 30 percent which means you are going to get faster responses and pay less for cost because you're using less tokens.

19:06It comes with a 256 ,000 context window and it has a 400 million parameter vision encoder for multi-model, meaning you can input text, images, and videos in order to get it to write the code that you need and it will understand all these inputs equally well. This model is also significantly cheaper than the competition. You can use it via Kimi Code on their platform with plans from$15 to$159 per month, or you can use it on the API where the billing is 95 cents for a million input tokens and$4 for a million output tokens. This is not an order of magnitude cheaper than the Western models, but it is still significantly cheaper than the Western competitions.

19:47So a lot is happening from a race perspective, but now let's talk a little bit about data security that we mentioned already. As I mentioned, the US government forced Anthropic to pull back its latest models, saying they are a real threat to national security. In addition, we know that Anthropic themselves said similar things about their model while developing Operation Glasswing, where they released it to a few companies to help patch the vulnerabilities that the model can find. And we've heard Dario Amadei and Demis Asabis both saying that we're running too fast and they wish we could slow down.

20:21The G7 summit has taken place last week in France, And all the big players were there, including Dario Amadei and Demis Asabis and Sam Altman and a few other players together with government leaders from the top seven countries in the world. And there were a clear call for international collaboration in order to define specific standards and safety measures for AI security. the two leading voices were Dario and Demis Assabis saying that they really think there needs to be a US-led international collaboration to monitor and define how to progress AI development in a safer way. Dario also spoke about how to exclude China from critical hardware and chip components as part of the effort to keep the world safe in a future where AI is so powerful.

21:17He said that multiple times before, this is not new. He also emphasized that there needs to be a cooperation to address AI risks in cyber, bioterrorism, and intelligence sectors. Sam Altman pushed for a similar thing, and he said, and I'm quoting, an international forum for discussion that establishes globally accepted standards for testing, provides experts an impartial analysis of capabilities and risks and serves as a venue for cooperation among nations. So not too specific, but definitely in the same direction. From a government collaboration perspective, French President Emmanuel Macron urged the U.S.

21:54not to monopolize cutting-edge AI, basically criticizing the export control on anthropic models as a strictly nationalist reaction. He also discussed the trusted partner scheme for non-US nations to access advanced US models, basically saying we are behind, we understand we are behind, but we don't want to be left completely behind. So please give us access to the models that you develop, even if you think that we shouldn't get access to them. Again, I don't know how this is going to evolve. I said that multiple times on this show. I truly, truly hope, pray, call it whatever you want to call it, that there will become a large international collaboration involving the leading development labs and governments and academia to figure out how to progress AI in a safe way for humanity.

22:39And I'm not just talking about cybersecurity safe. I'm talking about all the big issues. What happens if there's 30 % unemployment five years from now? How do we handle that? What happens to money if it becomes a bigger issue from a cash availability perspective? What happens to so many other aspects? Education that we take for granted right now, all of that has to change. And I think an international group that focuses on that with a lot of resources and a lot of smart people is the right way forward. Staying on the security of AI side of things, OpenAI received subpoenas from multiple states attorney generals investigating user safety concerns with Chachupiti.

23:17This comes just a few days after the company filed for its highly anticipated public offering. Many of these cases are known. Such a Canadian lawsuit alleges Chachupiti encouraged her daughter suicide decision. Florida's attorney general sued after two separate shooting cases where alleged gunmen consulted with Chachupiti during the crime's planning. OpenAI also faces scrutiny over how it uses health data and personal information, plus allegations that Chachupiti allegedly offered encouraging words to users considering self-harm and criminal acts. So it doesn't look great from OpenZEI's perspective, especially when they're coming into this, again, highly anticipated IPO.

24:02That being said, I don't think that is going to be significant enough to slow the IPO down or reduce its probably extremely high valuation. I don't think that Chachapiti from that perspective is different than most other models. And going back to my previous point, I think we need an international government intervention in order to verify specific aspects and capabilities of these models before they can be released to the public. That raises an even bigger picture on what happens with the open source models that, as we mentioned earlier, are taking a bigger share and providing better and better capabilities that are now in many cases aligned with the most advanced closed source models.

24:43Now, the third topic I want to give you a quick brief about is new developments in the agentic universe. The first news comes from Databricks, who open sources Omnigent, which is an open source meta harness that standardizes how multiple AI agent frameworks from cloud code, codex, pi, open AI agents, custom SDKs work together. So the tool is built to address the aspect of you don't have to be tied into one platform. You can develop multiple agents in multiple tools where they have better capabilities and use Omnigent in order to create a unified interface to allow to develop governance, composition, coordination, and collaboration across all the existing harnesses and build a more robust, more capable, more flexible overall solution.

25:31In addition to connecting between the different platforms, it also connecting across all the different interface capabilities. So a single session can synchronize across terminal, web user interface, desktop, mobile, and APIs simultaneously, replacing the need to jump and copy and paste between multiple tools that you're using at the same time. Omnigent also comes with contextual policies at the meta-harness layer rather than through specific prompts, which means you can pause an agent after a special specific spend or when the output is not aligned with whatever guidelines you define, regardless of which platform it is coming from, giving companies who are doing this already, meaning running and developing agents across multiple platforms, a lot more controls, both from a spending capability as well as from a data security perspective.

26:22And I think this is going to be a highly needed capability in the very near future for more and more companies. And I have a feeling Omnigent is not the last tool that is doing this that we're going to see. I will be extremely surprised if the labs themselves do not allow to use some of their platforms to manage agents from other solutions because it will become a necessity. and then not allowing this will just mean that people will go to third-party platforms versus using your platform, which means you're using control, knowledge, data, and a lot of other things. So I don't think this is the last one, but it is definitely a very interesting step in a new direction that was not available as a unified tool before.

27:03Staying in the same topic, Vercel just launched EVE, which is an open source agent framework that is designed to simplify the building, running, and scaling of AI agents by integrating production-ready features directly into the framework. Now, as part of this release, they shared something that is not surprising and yet very interesting with how fast it's evolving. AI agents now account for more than half of all commits on Verso's platform. This was less than 3 % at the beginning of this year. So in the beginning of 2026, Verso saw practically no agentic commits, and now it takes more than half of the commits on the platform.

27:42Now, Eve includes basically all the components you would expect to allow people to develop and run agents, so execution capabilities, sandbox compute for security, human in the loop approvals, sub-agents for delegation, and a built-in evals and testing quality assurance level built into the platform. Now, this entire framework allows agents to be deployed as regular Verso projects supporting various channels such as Slack, GitHub, Snowflake, Salesforce, Notion, Linear, Discord, Teams, Telegram, Twilio, and many others, which again, I think a lot of people that are already using Vercel, and there are millions and millions of people doing it, I'm using it for several of my deployments, will see a benefit in.

28:23Now, Vercel themselves have been using the platform now internally. They're running over 100 agents on the EVE platform, which include things such such as data analysis, which runs more than 30 ,000 questions every single month, an autonomous sales development representative that is returning 32x its annual cost of$5 ,000, a sales cockpit, a support engineer that is now solving 92 % of internal tickets on its own, content agents, etc. So all of these are developed and running inside the Vercel environment, providing very obvious value to anybody who is using them. Staying on the topic of features that enable more advanced functionality, but this one is relevant to a lot more people, is that OpenAI has rolled out a significant update to the ChatGPT ability to schedule tasks.