Claude Code now has a mode for managing multiple agents at once, which should reduce the number of open terminals many times over - and developers building orchestrators on top of Claude Code are probably sweating.
On top of that, the Anthropic team completely copied Codex’s /goal mode: in this mode, the model does not stop until it reaches the goal.
On top of that, the Anthropic team completely copied Codex’s /goal mode: in this mode, the model does not stop until it reaches the goal.
Fresh stats also came out today from Ramp, a company that manages corporate spending and issues cards, for example for employees on business trips
According to their data, 50% of U.S. companies now pay for AI subscriptions, and for the first time Anthropic’s share has exceeded OpenAI’s, which has barely grown for almost a year.
The most common criticism of this analysis is that many large companies do not use Ramp, so the statistics are not exactly precise.
But it is what it is - and in response, Uncle Sam announced that companies can try Codex for free for two months. You can try filling out the form here; I don’t see any country restrictions, so maybe it works outside the U.S. too.
🥊 The battle for the market is on.
According to their data, 50% of U.S. companies now pay for AI subscriptions, and for the first time Anthropic’s share has exceeded OpenAI’s, which has barely grown for almost a year.
The most common criticism of this analysis is that many large companies do not use Ramp, so the statistics are not exactly precise.
But it is what it is - and in response, Uncle Sam announced that companies can try Codex for free for two months. You can try filling out the form here; I don’t see any country restrictions, so maybe it works outside the U.S. too.
🥊 The battle for the market is on.
Anthropic will give all subscribers Claude credits for third-party apps
Starting June 15, subscribers will receive credits equal to the price of their subscription - $20, $100, or $200, depending on the subscription tier. These credits will not affect the limits of the main subscription in any way.
They can be used for apps built on the Agent SDK, such as OpenClaw or custom tools using Claude.
That’s what life-giving SpaceX compute does.
Starting June 15, subscribers will receive credits equal to the price of their subscription - $20, $100, or $200, depending on the subscription tier. These credits will not affect the limits of the main subscription in any way.
They can be used for apps built on the Agent SDK, such as OpenClaw or custom tools using Claude.
That’s what life-giving SpaceX compute does.
If you think AI is already everywhere, here’s a chart to think about
Red — people who use agents;
Yellow — people who pay for chatbots;
Green — people who have ever used free tiers.
The rest are people who have never used AI at all.
Looks like the technology is still at a pretty early stage of adoption. And even now, we already don’t have enough compute for inference.
Red — people who use agents;
Yellow — people who pay for chatbots;
Green — people who have ever used free tiers.
The rest are people who have never used AI at all.
Looks like the technology is still at a pretty early stage of adoption. And even now, we already don’t have enough compute for inference.
xAI is buying up tax returns to train Grok
In March of this year, xAI asked employees, as well as their friends and relatives, to provide their tax returns for training Grok. In exchange, they were promised $420.
Now employees are complaining that two months later, no one has been paid. The company says the person responsible for the program no longer works there.
https://www.bloomberg.com/news/articles/2026-05-18/musk-s-xai-promised-staff-420-for-their-tax-returns-hasn-t-paid
In March of this year, xAI asked employees, as well as their friends and relatives, to provide their tax returns for training Grok. In exchange, they were promised $420.
Now employees are complaining that two months later, no one has been paid. The company says the person responsible for the program no longer works there.
https://www.bloomberg.com/news/articles/2026-05-18/musk-s-xai-promised-staff-420-for-their-tax-returns-hasn-t-paid
Elon Musk lost his case against OpenAI
A nine-member jury in a federal court in Northern California reached a unanimous verdict: all three main claims in Elon Musk’s lawsuit against Sam Altman and Greg Brockman — breach of a charitable trust, aiding and abetting that breach, and unjust enrichment — were rejected because the statute of limitations had expired.
In other words, Altman and Brockman were not technically cleared of wrongdoing, but Musk still walks away with nothing. There will be no compensation and no restructuring of OpenAI, which he had been seeking.
The judge agreed with the jury’s verdict and formally dismissed all the claims.
https://techcrunch.com/2026/05/18/elon-musk-has-lost-his-lawsuit-against-sam-altman-and-openai/
A nine-member jury in a federal court in Northern California reached a unanimous verdict: all three main claims in Elon Musk’s lawsuit against Sam Altman and Greg Brockman — breach of a charitable trust, aiding and abetting that breach, and unjust enrichment — were rejected because the statute of limitations had expired.
In other words, Altman and Brockman were not technically cleared of wrongdoing, but Musk still walks away with nothing. There will be no compensation and no restructuring of OpenAI, which he had been seeking.
The judge agreed with the jury’s verdict and formally dismissed all the claims.
https://techcrunch.com/2026/05/18/elon-musk-has-lost-his-lawsuit-against-sam-altman-and-openai/
Google is getting ready to release an insanely cool video model
Their annual Google I/O event is happening today, and they’re already dropping spoilers left and right. Judging by the looks of it, we’re in for a powerful omni model — possibly state-of-the-art in video generation.
For example, videos like these were posted ahead of the event by Google CEO Sundar Pichai and Logan Kilpatrick, the main source of Gemini leaks.
Can’t wait.
Their annual Google I/O event is happening today, and they’re already dropping spoilers left and right. Judging by the looks of it, we’re in for a powerful omni model — possibly state-of-the-art in video generation.
For example, videos like these were posted ahead of the event by Google CEO Sundar Pichai and Logan Kilpatrick, the main source of Gemini leaks.
Can’t wait.
Cursor has released Composer 2.5
Over the past two months, the model has improved noticeably on benchmarks while using the same K2.5 base. The number of synthetic tasks used to train the model was increased 25-fold. This is Cursor’s first model trained in SpaceXAI data centers, and they are already jointly training a much larger model using 10 times more compute. Most likely, the results of that model will determine whether SpaceX acquires Cursor or not.
With the release of the new model, they doubled the price of fast mode, which is enabled by default. It now costs $3/$15 per million tokens, matching the price of Sonnet. The price of the regular mode has not changed — $0.50/$2.50.
Over the past two months, the model has improved noticeably on benchmarks while using the same K2.5 base. The number of synthetic tasks used to train the model was increased 25-fold. This is Cursor’s first model trained in SpaceXAI data centers, and they are already jointly training a much larger model using 10 times more compute. Most likely, the results of that model will determine whether SpaceX acquires Cursor or not.
With the release of the new model, they doubled the price of fast mode, which is enabled by default. It now costs $3/$15 per million tokens, matching the price of Sonnet. The price of the regular mode has not changed — $0.50/$2.50.
OpenAI co-founder Andrej Karpathy now works at Anthropic
Recently, Karpathy had been making educational AI content on his YouTube channel. He promised not to give up that work.
TechCrunch learned that at Anthropic, Karpathy and his new team will focus on using Claude to speed up the pretraining of new models.
https://techcrunch.com/2026/05/19/openai-co-founder-andrej-karpathy-joins-anthropics-pre-training-team/
Recently, Karpathy had been making educational AI content on his YouTube channel. He promised not to give up that work.
TechCrunch learned that at Anthropic, Karpathy and his new team will focus on using Claude to speed up the pretraining of new models.
https://techcrunch.com/2026/05/19/openai-co-founder-andrej-karpathy-joins-anthropics-pre-training-team/
So, Google I/O. The main releases:
1️⃣ Gemini 3.5 Flash. According to benchmarks, it beats Gemini 3.1 Pro in coding, agentic tasks, and multimodality, while generating at just under 300 tokens per second. That is noticeably faster than GPT-5.4 mini and Claude Haiku, and roughly 4 times faster than Opus, Sonnet, and GPT-5.5. They also showed a version during the presentation that outputs almost 1,500 tokens per second. The price is also much more affordable than the Pro versions, GPT, and Opus.
Gemini 3.5 Pro was promised for next month. Gemini 3.5 Flash is already available to try in Antigravity.
2️⃣ Gemini Omni. As Google put it, this is “the first step toward models that can generate anything from anything.” In its current version, Omni is limited to video generation. It can take text, images, video, and voice as input, and outputs video. The generation quality is competitive, the physics holds up, and the photorealism is solid. You can try it in the Gemini app.
3️⃣ Antigravity 2.0. They added a bunch of features for orchestrating parallel agents and subagents, background tasks, and native voice command support. They released Antigravity CLI, replacing Gemini CLI, and the Antigravity SDK for building custom agents. During the presentation, they showed the system building a working operating system that can run Doom. It took less than 12 hours, 96 agents, and under $1,000.
4️⃣ And a few smaller things:
- they completely redesigned the Gemini app interface, so answers now look more like interactive pages;
- AI Plus, Pro, and Ultra subscribers now get Daily Brief, a personalized daily digest based on Gmail, Calendar, and tasks;
- Ultra now includes Gemini Spark, a personal agent that keeps running even when your laptop is off and completes tasks for you. For now, it only integrates with Google tools, but they promised to add more soon;
- the Google AI Ultra subscription now starts at $100/month, down from $250, while the $250 tier has dropped to $200 with no cuts to features or limits;
- they announced smart glasses and new Googlebooks laptops.
1️⃣ Gemini 3.5 Flash. According to benchmarks, it beats Gemini 3.1 Pro in coding, agentic tasks, and multimodality, while generating at just under 300 tokens per second. That is noticeably faster than GPT-5.4 mini and Claude Haiku, and roughly 4 times faster than Opus, Sonnet, and GPT-5.5. They also showed a version during the presentation that outputs almost 1,500 tokens per second. The price is also much more affordable than the Pro versions, GPT, and Opus.
Gemini 3.5 Pro was promised for next month. Gemini 3.5 Flash is already available to try in Antigravity.
2️⃣ Gemini Omni. As Google put it, this is “the first step toward models that can generate anything from anything.” In its current version, Omni is limited to video generation. It can take text, images, video, and voice as input, and outputs video. The generation quality is competitive, the physics holds up, and the photorealism is solid. You can try it in the Gemini app.
3️⃣ Antigravity 2.0. They added a bunch of features for orchestrating parallel agents and subagents, background tasks, and native voice command support. They released Antigravity CLI, replacing Gemini CLI, and the Antigravity SDK for building custom agents. During the presentation, they showed the system building a working operating system that can run Doom. It took less than 12 hours, 96 agents, and under $1,000.
4️⃣ And a few smaller things:
- they completely redesigned the Gemini app interface, so answers now look more like interactive pages;
- AI Plus, Pro, and Ultra subscribers now get Daily Brief, a personalized daily digest based on Gmail, Calendar, and tasks;
- Ultra now includes Gemini Spark, a personal agent that keeps running even when your laptop is off and completes tasks for you. For now, it only integrates with Google tools, but they promised to add more soon;
- the Google AI Ultra subscription now starts at $100/month, down from $250, while the $250 tier has dropped to $200 with no cuts to features or limits;
- they announced smart glasses and new Googlebooks laptops.
GitHub has been hacked
A breach of roughly 3,800 internal GitHub repositories has been confirmed. User repositories were not affected.
The attack happened through a GitHub employee’s device that had an infected VS Code extension installed.
Meanwhile, the hackers are selling GitHub’s source code for $50,000. If they do not find a buyer, they say they will publish it for free.
https://x.com/github/status/2056949168208552080?s=20
A breach of roughly 3,800 internal GitHub repositories has been confirmed. User repositories were not affected.
The attack happened through a GitHub employee’s device that had an infected VS Code extension installed.
Meanwhile, the hackers are selling GitHub’s source code for $50,000. If they do not find a buyer, they say they will publish it for free.
https://x.com/github/status/2056949168208552080?s=20
Altman offered $2 million in tokens to every startup in the current YC batch in exchange for equity
A completely logical move. It’s surprising this did not become the market standard about three years ago.
Investing in every YC startup across the board is, in principle, a strategy that can work: you can make money on it, or at least not lose all that much. Investing in tokens means baking in a 10x margin from the start. And on top of that, you get to lock them into your ecosystem for the future with what is basically free money.
https://x.com/sama/status/2056933166875857290
A completely logical move. It’s surprising this did not become the market standard about three years ago.
Investing in every YC startup across the board is, in principle, a strategy that can work: you can make money on it, or at least not lose all that much. Investing in tokens means baking in a 10x margin from the start. And on top of that, you get to lock them into your ecosystem for the future with what is basically free money.
https://x.com/sama/status/2056933166875857290
It suddenly turned out that DeepMind founder Demis Hassabis has a stake in Anthropic
The Financial Times conducted an investigation and found that he was one of the startup’s early so-called angel investors and had invested in the company secretly.
Now Demis and the Amodeis are practically among each other’s biggest competitors.
Despite that, let’s remember that Google recently signed a $10 billion investment agreement with Anthropic. Could Hassabis have had a hand in it?
https://www.ft.com/content/8f2a529e-7a1b-4d8e-95be-338d0c4c98f5
The Financial Times conducted an investigation and found that he was one of the startup’s early so-called angel investors and had invested in the company secretly.
Now Demis and the Amodeis are practically among each other’s biggest competitors.
Despite that, let’s remember that Google recently signed a $10 billion investment agreement with Anthropic. Could Hassabis have had a hand in it?
https://www.ft.com/content/8f2a529e-7a1b-4d8e-95be-338d0c4c98f5
Anthropic expects its first profitable quarter
Anthropic told investors that it may reach operating profitability for the first time in the second quarter. Quarterly revenue is expected to more than double to around $10.9 billion.
At the same time, it emerged that Anthropic will pay xAI $1.25 billion a month for compute capacity. The deal concerns the Colossus 1 data center near Memphis. Anthropic bought out its entire 300-megawatt capacity.
The contract runs until May 2029 and could bring xAI more than $40 billion. For Musk, this is a good way to put to use capacity that appears to have turned out excessive for Grok.
https://www.wsj.com/tech/ai/mind-blowing-growth-is-about-to-propel-anthropic-into-its-first-profitable-quarter-7edbf2f4?st=HwPZX6&reflink=desktopwebshare_permalink
Anthropic told investors that it may reach operating profitability for the first time in the second quarter. Quarterly revenue is expected to more than double to around $10.9 billion.
At the same time, it emerged that Anthropic will pay xAI $1.25 billion a month for compute capacity. The deal concerns the Colossus 1 data center near Memphis. Anthropic bought out its entire 300-megawatt capacity.
The contract runs until May 2029 and could bring xAI more than $40 billion. For Musk, this is a good way to put to use capacity that appears to have turned out excessive for Grok.
https://www.wsj.com/tech/ai/mind-blowing-growth-is-about-to-propel-anthropic-into-its-first-profitable-quarter-7edbf2f4?st=HwPZX6&reflink=desktopwebshare_permalink
OpenAI is starting the official process of going public
The WSJ writes that the company plans to file with regulators literally today or tomorrow. Apparently, everything was already ready, and they were just waiting for the lawsuit with Musk to end.
Once the filing is submitted, it will be reviewed for several months, so the IPO could happen in the fourth quarter of 2026 - and it may very well become one of the largest in history.
The WSJ writes that the company plans to file with regulators literally today or tomorrow. Apparently, everything was already ready, and they were just waiting for the lawsuit with Musk to end.
Once the filing is submitted, it will be reviewed for several months, so the IPO could happen in the fourth quarter of 2026 - and it may very well become one of the largest in history.
You can now vibe-code Android apps in Google AI Studio
You describe your idea, and AI Studio generates native Kotlin code. The app can be tested right away in a browser-based emulator and installed on a device.
https://blog.google/innovation-and-ai/technology/developers-tools/google-ai-studio-io-2026/
You describe your idea, and AI Studio generates native Kotlin code. The app can be tested right away in a browser-based emulator and installed on a device.
https://blog.google/innovation-and-ai/technology/developers-tools/google-ai-studio-io-2026/