Claude

Europe forces Google to open Android to ChatGPT and Claude, but the blow is that it will have to give up its search engine data to the competition

by

The European Commission has ordered Google to open two of its greatest competitive advantages– Android’s access to rival AI assistants and the search data that feeds Gemini. They are two binding measures under the Digital Markets Act (DMA)adopted on July 16. Android must be open by July 2027, search data by January 2027, and no resources are stopping those deadlines. In detail. The first measure affects how a rival assistant can live inside Android. Right now, Gemini has privileges that no competitor shares: Keyword activation at the operating system level, not the application level. Access to the home button or navigation gesture as a gateway. Reading content on screen to respond with context. Executing tasks within other applications, such as sending an email or sharing a photo. Priority access to hardware and AI models built into the device. There are 11 points in total. Starting with the next major version of Android, ChatGPT, Claude or any other attendee will be able to claim the same treatment that only they have today Gemini. The second measure is the one that weighs the most for the business: the data of queries, clicks and results with which Google trains a good part of its LLM. From January 2027, rivals such as OpenAI will be able to access an anonymized version, under conditions that Brussels demands are fair and non-discriminatory. Google’s waterline. Between the lines. Google completed a few months ago the transition from Google Assistant to Gemini as Android’s default assistant, just before Brussels opened the file. The order arrived with Gemini already installed as a part of the system and not as just another application. Apple chose the opposite path: it negotiated before launching, proposed a monitored permit system and Brussels rejected it. Siri AI It still hasn’t reached the iPhone in the European Union. Google launched Gemini without asking permission and has discussed the conditions afterwards, with better results. Yes, but. Google, like Apple, alleges security and privacy reasons. Its head of global affairs, Kent Walker, has explained that opening screen reading and the execution of tasks between applications to any external service bypasses the filter that the manufacturers themselves apply today before granting these permissions. And now what. Google will appeal, but the appeal does not give it extra time: from the criteria established in the Apple case, the obligation is fulfilled while it is discussed whether it was legal to impose it. The United Kingdom and Japan, which are preparing their own digital competency frameworks, are presumably taking note of how “fair access” to a mobile is defined here. What Brussels sets about wake words or search data prices will not remain only in Europe. In Xataka | AI has destroyed traditional captchas. Google’s solution requires scanning a QR… and having its services installed Featured image | Xataka

Anthropic already had Claude writing code. Now he has put it in the laboratories

by

Anthropic had already placed Claude in one of the most everyday and valuable tasks in the technology industry: writing code. Now he wants to take it to more delicate terrain and with potentially much greater consequences: scientific work within laboratories. The company has introduced Claude Sciencea product designed to help researchers move between literature, data, specialized tools and computing resources. Claude to science. The key to Claude Science is not only that Anthropic has added more tools to Claudebut in the type of problem it is trying to solve. In science, a huge part of the work involves jumping between databases, files, code, figures, citations, and computing resources that rarely talk to each other comfortably. The company wants to integrate all this into a specific application, available from June 30, 2026 in beta for Pro users, MaxTeam and Enterprise on macOS and Linux. A category jump. Anthropic had already begun to bring Claude closer to scientific work last fall, when it launched connectors and functions under the umbrella of Claude for Life Sciences. This helped the model to relate better to software and scientific databases, but it still had a more limited scope. What is happening now goes one step further. Anthropic seems to want science to stop being just a use case and become a product line. Verifiable work. The promise of Claude Science is not limited to helping you write or summarize. Anthropic claims it can analyze scientific literature, execute multi-step investigations, generate figures and manuscripts, and allow the researcher to refine them iteratively. The most important part is how it leaves a trace: each result includes the code, the environment, and the message history that produced it. In addition, a review agent checks quotes and calculations, and can point out untraceable numbers or figures that do not match the code that generated them. Claude Science’s ambition might sound very broad, but his first steps have a fairly recognizable accent. Anthropic has prepared it with more than 60 capabilities and connectors targeting areas such as genomics, proteomics, structural biology, computational chemistry, and single-cell analysis. The computation, within the flow. Many investigations do not stop at reading articles or generating figures: they also require carrying out heavy work on machines prepared for it. Anthropic says Claude Science can help prepare those processes on the researcher’s laptop, on a Linux machine, on an HPC access node via SSH, or with on-demand computing in Modal. The company clarifies that the system writes a plan and asks permission before accessing new resources, so that the researcher can review or revoke decisions. It also states that large or sensitive data can remain in the lab infrastructure, sending Claude only the context necessary for each step of the analysis. Anthropic accompanies the launch with examples. Manifold Bio, dedicated to the design of drugs aimed at specific tissues, used Claude Science to propose targets in its experiments, evaluating surface expression, cell trafficking and safety according to the company’s own criteria. The Allen Institute used it to build a computational review template with about 20 custom skills, capable of reading thousands of articles and organizing findings into an evidence base. And at UCSF, epidemiologist Stephen Francis says the tool sped up glioma analysis to about one-tenth the time before, with results independently validated by his group. Images | Anthropic In Xataka | South Korea has a plan to dominate in memory chips and robotics. One of a billion dollars

What is Eustella, prices and what you can do with the new European alternative to Claude, Gemini or ChatGPT but with agents

by

Let’s tell you What is Eustella and how does it work? this European alternative to OpenClaw. It is an artificial intelligence assistant for Android and iOS, which is promoted with privacy as its flag by having all the data hosted on servers in the United States. We are going to start the article by explaining what this AI is and who is behind it. Then we will tell you all the things you can do with Eustellaand we will finish by telling you what the prices of their payment plans are. What is Eustella and who is behind it? Eustella is an artificial intelligence personal assistant created by a European company. Created in Europe for Europeans, says the Vienna-based company newsrooms.ai, which is responsible for this project. In essence, it is an AI assistant with agent capabilitiesso it offers several possibilities. This means that you can do two things with this AI. First of all, you can ask him any type of questions as you would with Claude, ChatGPT either Gemini. But secondly you can also ask him to perform tasks for you on your mobile device. Its big argument is that you maintain full control over your information, and that no profiles are built or AIs are trained with it. In addition to having been developed by an Austrian company, the data is hosted entirely in servers located in Europewhich are managed by the German company IONOS CLOUD. Of course, not even IONOS has access to user data, since the information is stored encrypted. Under the hood, we are going to find several artificial intelligence models. On the one hand we have European Mistral modelsbut also other open weight models such as Qwen 3.5/3.7 from Alibaba, Gemma 4 from Google, gpt-oss-120b from OpenAI and Flux from Black Forest Labs. Non-European models use them because they are open source and can be hosted in Europeno matter where they were trained. With this, Eustella does not depend on the closed APIs of the United States, and in fact explicitly avoids the use of proprietary APIs. In addition, developers intend to continue maintaining flexibility by integrating new open weight models as soon as they offer better results. How it works and what it can do You can use Eustella both in the browser and on the mobile. You can register and access the website eustella.com or access your app for Android. An iOS app will also be released soon. Therefore, even for its functions of AI agentusing it is as easy as installing an app. This AI seeks ease of use above all else. You will not have to install anything additional, you do not have to select or configure models or servers. You simply write and the bot will respond. As for its capabilities, the first thing is that Eustella works as an artificial intelligence chatbot. You can ask him any question and he will answer you by searching for information in the knowledge of his models or on the Internet. Furthermore, it also has deep investigation featuresadvanced search or answer specific questions about Eustella itself. Another of its functions is multimedia creation. Can generate images from scratch or modify existing ones. For that, it also has a vision mode, allowing you upload images to analyze them or extract text from them. It also has office functionsbeing able to create, read and edit Word, PowerPoint or Excel documents. It offers professional writing or even file analysis, being able to read and extract content from PDF files or other text documents. Eustella also allows you schedule remindersmanage notes and create projects to centralize queries on the same topic there. You can also manage your memories, or even create a companiona kind of virtual pet with its own personality to interact with and ask for things. Another of the great strengths is that you have several AI agents that you can use. Each of them perform specific automations, such as helping you plan a trip, having a morning news summary, finding music, conducting research, helping you learn, etc. EUstella also allows you to create your own agents, group chat and more. Additionally, it also has integrations in place, such as displaying real-time cryptocurrency prices with Bitpanda or a price comparator from Geizhals. In the coming months there will be new integrations focused on maps, travel, navigation, shopping, education and lifestyle. Price of Eustella subscriptions You can use Eustella for free, although You also have three payment plans. They are the following, and with the following characteristics: Comet (€5.99/month): Gives you more tokens and more Eustella usage. You can do more projects and advanced search, document creation, image generation and European data sovereignty features are added. Star (€17.99/month): Everything Comet offers but with usage limits 3 times higher. It also allows you to create more projects, more image generation and you can schedule agents at specific times. Cosmos (€89.99/month): Everything Star offers, but with 15 times the usage limit. You can also create more AI agents, have unlimited projects, and European priority for support. In Xataka Basics | Artificial intelligence on your PC: the best free tools to install AI models like DeepSeek, Llama, Mistral, Gemma and more

Someone has created a PS1 programming environment with Claude. And he has started programming games

by

Manny has been wanting this software to exist for 20 years. Security engineer by day, indie developer by night, he grew up with a PlayStation and never stopped thinking about making games for it. When he finally got around to trying, the barrier wasn’t the complexity of the hardware: it was that the tools available were in C++ and getting them started was too difficult. Their solution was the most radical possible: build the tools from scratch. Have your own emulator. The result is called PSoXide. According to describes Manny himselfis a complete development stack for PS1 written entirely in Rust: emulator, SDK, engine, level editor and pipeline disk. And Claude is built into the emulator as part of the architecture. “This would have been completely impossible without the LLMs,” says Manny, and it is not an exaggeration: PSoXide is a highly complex project, which would have required a team behind it, but which he has been able to emulate on his own thanks to AI. How it works. The emulator implements the complete machine: the R3000A CPU, the GPU, the GTE, the CD-ROM, the DMA, the SPU, the timers and the MDEC. As detailed in the Project READMEManny built a suite of tests with more than 65 ROMs, and then made a test CD to run it on a real PS1 and be able to check if the emulator was imitating the original machine one hundred percent. The SDK, in short, is very complete: it communicates directly with the hardware instead of going through the console’s BIOS for performance reasons. And it has dedicated modules for GPU, GTE, controllers, SPU audio with ADPCM, CD-ROM and ISO9660. And memory cards, of course. Without Claude it can’t be done. Manny says the part he’s most excited about about the editor is that it “wires the emulator directly to Claude”: there’s an MCP server that exposes about 25 endpoints of debugso the model can inspect the state of the CPU, VRAM, and registers as they work. “Debugging a PS1 game by simply asking Claude to inspect the hardware is like a wet dream,” he says. When something goes wrong in a game’s behavior, Manny can ask Claude to directly read the logs from the emulated hardware and diagnose what’s going on. That design is a direct consequence of how the project uses AI: PSoXide was developed “with substantial assistance from AI, with a human leading the architecture, the debugging and hardware verification.” That is, and this is also important and makes clear certain preconceptions about how AI can replace programmers: he directs, AI executes. Being “a software engineer in the pre-AI era means that in the long term I can guide the models in a way that the code remains stable and expandable.” Manny: Origins. According to your profile on itch.ioManny started programming with the PICO-8, whose limitations (small resolution, small cartridges) taught him what restriction can do for creativity. Then he built BONNIE-32its own console in Rust, with software renderer and integrated tools. In parallel, in February 2026, it published an experimental alpha version of a port of ‘Zelda: Ocarina of Time’…for PS1. He made it playable, with 3D movement and a camera controllable with the sticks, and conceived it as a learning tool for real hardware. The catalogue. The first game published on PSoXide is the ‘Celeste Classic Collection PSX‘: Maddy Thorson and Noel Berry’s two PICO-8 games ‘Celeste Classic’ and ‘Celeste 2: Lani’s Trek’, rewritten in Rust on the PSoXide SDK. Be careful, we are not looking at a PICO-8 emulator running on the PS1, but rather native code: an executable of approximately 468 KB, graphics with 4 bits per pixel textures, real-time synthesized audio with 8 voices… This beauty runs at 60 frames per second and is downloaded on itch.io at a free price. And after that, a much more ambitious project: Manny is working with a 3D artist on an original game for PS1, a souls-like with three-dimensional graphics for the console. Or as Manny says, “aimed at something you really care about, AI doesn’t replace your creativity, it unlocks it.” In Xataka | Claude Code is being the big favorite among programmers. So much so that he already signs 4% of everything that is uploaded to GitHub

Claude Guillemot, the co-founder of Ubisoft who turned a store in Brittany into a video game empire, dies

by

Claude Guillemot, one of the five brothers who founded Ubisoft in 1986 and until now current executive vice president of operations of the company, died yesterday, June 19, 2026 at the age of 69 when his Cessna 421 plane crashed in a field near the La Baule airfield, in the department of Loire-Atlantique, as reported by France 3. The other fatality is a flight instructor from Rennes whose identity has not yet been confirmed. Ubisoft’s silent architect. Although his brother Yves Guillemot is the most visible face of the company as CEO, Claude’s role in the birth of Ubisoft was essential: according to 3DJuegoswas the one who opened Guillemot Informatique in Carentoir in 1984, the original computer store that served as the base for Ubisoft that would be born just a couple of years later. Until his death he was an active member of the Board of Directors and, according to your company filebrought to the board his international experience in Asia and his in-depth knowledge of video game hardware and distribution technologies. He was also CEO of Guillemot Corporation since 1997, a company specialized in video game peripherals under the Thrustmaster and Hercules brands, present in more than 140 countries. At Ubisoft’s worst moment. Guillemot’s death hits a company that is going through a difficult time: Ubisoft has recorded record losses of 1.3 billion euros in the 2025-26 fiscal year, with a drop in income of 17.4%. The crisis has specific causes: has canceled six gamesincluding the remake of ‘Prince of Persia: The Sands of Time‘, and delayed another seven, assuming as lost the 650 million already spent on development. To achieve liquidity, has transferred 25% of Vantage Studios to Tencent for 1,160 million euros. The company foresees that the next year will also be in the red, with an additional estimated drop in sales of between 8% and 9% and does not expect to return to profitability until the 2027-28 financial year. Yves Guillemot himself advertisement in January 2026 a large-scale “reset”, describing the company’s momentum as an opportunity to return to sustainable growth. The end of an era. In the absence of the Bureau d’Enquêtes et d’Analyses completing the investigation into the accident, what is already definitive is the loss of one of the five architects of the only major European video game multinational capable of competing with the major North American and Japanese studios for four decades. The Guillemot family controls around 14% of Ubisoft’s capital through Guillemot Brothers SE (according to data from Euronext Paris), and the reorganization of that control structure will be one of the first issues that the market will monitor in the coming weeks. In Xataka | What’s happening with Ubisoft: after canceling six games and adjusting its structure, this is the plan of the great French studio In Xataka | A single programmer, simple mechanics, crappy graphics and Paint interface. And he has earned ten million in a week Cover | Shuichi Aizawa

The US did not turn off Claude Mythos for fear of a jailbreak, it did so because it suspected someone was watching. That someone is China

by

Last Saturday, the White House ordered Anthropic to remove access to Fable 5 and Mythos to any foreign person, regardless of whether they were inside or outside the United States. This order supposedly came after a tip that there was a way to jailbreak the model and bypass safeguards, but according to Wired it came at a time when Washington was already uneasy for another reason. South Korea. They tell it in Wired. According to sources close to the government, the notice of the possible jailbreak came at a time when the US government had already raised its eyebrows over another issue. It turns out that one of the large companies that had access to Mythosthe original model without the safeguards, was SK Telecom, the largest telephone operator in South Korea. The concern, according to these sources, arises from the alleged ties of this operator with China. The logic would have been something like: if SK Telecom has access, China has access. Furthermore, they have told us that there are vulnerabilities. Let them turn it off now. Why is it important. Until now it was believed that the reason behind the decision was the warning about a possible jailbreak in Fable, which was the “trimmed” version of Mythos that was released to the public. It was Amazon’s own CEO, Andy Jassy, ​​who called the treasury secretary to warn him that they had found vulnerabilities in Fable. In this version of the story, the government’s concern was that anyone would bypass Fable’s safeguards and use it for evil. However, with this new information things change: it is no longer just any person, but China, its great enemy in the technological war, which could be looking into the guts of Mythos. They already had them in their sights. As we said, SK Telecom was one of the 150 companies chosen to test Mythos within the framework of what is known as Project Glasswing. According to internal sources consulted by Wired, the US government requested that Anthropic revoke their access earlier this month, to which Anthropic agreed. It seemed like the thing was going to stay there, but when Fable was released to the public and learned that it had a vulnerability, the White House reacted. SK Telecom, Anthropic and China. The relationship between Anthropic and SK Telecom dates back to 2023, when They invested 100 million dollars in the AI ​​startup for the development of AI models focused on telecommunications. Currently, SK Telecom barely has a presence in China (They only have seven employees and their billing is testimonial), but in the past they had a very close relationship with China Unicom, creating a joint venture and investing up to $1 billion in the Chinese operator. Although already They sold their shares in 2009SK Telecom is part of the SK Group conglomerate, which has other businesses beyond telecommunications, such as semiconductors and energy, these with alliances in China. SK Telecom has denied maintaining ties with China. Previous tensions. All this happens at a time when the relationship between the US government and Anthropic is not going through its best moment. In March of this year they starred in a whole soap opera Anthropic’s refusal to allow military use of its models without safeguards. They ended up on what is known as the black list.Anthropic took the US to court and now we have a new season of the soap opera with the whole Mythos thing. Image | Xataka with Magnific In Xataka | China has two ideas to win the AI ​​race: invest a fortune and leave NVIDIA with almost no margin

The United States presses the fear button with Claude Mythos and Fable 5. It is a blackout with a clear beneficiary: China

by

The United States has turned off the tap most powerful, expensive and advanced AI model ever released by Anthropic, but it has done so in a disconcerting move and with small print. Yesterday, June 12, the White House ordered Anthropic to immediately shut down Fable 5 and Mythos 5 for anyone who is not a US citizen. The order arrived without giving too many explanations about its duration or technical justification, according to Anthropic’s public statement. This government intervention marks a before and after in the modern technology sector: they are not restricting the sale of chips (a common practice), but vetoing access to general access software as a matter of national cybersecurity. What’s happening with Fable 5 and Mythos 5. The US government has issued an export control directive suspending all access to Fable 5 and Mythos 5 for anyone who does not have US citizenship, whether inside or outside the United States. This includes non-U.S. Anthropic personnel. In order to adhere to the directive that came into effect immediately, the company led by Dario Amodei has had to deactivate both models for everyone. At the time of publishing this article, the notice appears when trying to access it. According to Anthropic, the government believes it has discovered a way to jailbreak the model. The company does not agree: after reviewing the demo of this technique to deceive the model, it concludes that the vulnerabilities found are minor, were already located and applicable to other models on the market and can be identified without the need for any circumvention technique. What you see right now if you enter Claude Why is it important. Because Mythos It is a specifically designed model for cybersecurity and its blackout affects both commercial users and those entities in charge of active defense management that are using it. With this measure, Washington turns its cutting-edge AI software into a national security asset. It is worth remembering that Anthropic is already on the blacklist of the Pentagon, so he considers it too dangerous for his Government to use. It is also now too dangerous for foreigners to use. Besides, as Anthropic points outsets a dangerous precedent: if the criterion for withdrawing a model is that someone finds a vulnerability, in practice none will be able to be launched because the company that releases software without a single failure raises its hand. That is to say, if the White House makes this exceptional movement the norm, it will be a shot in the foot to develop cutting-edge models: slower launches due to this “impossible” debugging process, personnel decimated by not being able to count on foreign specialists and if they cannot be marketed abroad there will be less income. Context. This order is the latest chapter in a soap opera of disagreements between Washington and Anthropic that dates back to the beginning of the year. In March, the Pentagon considered that the company was a “supply chain risk.” In its statement, Anthropic has verified that that level of capacity that the government identified as dangerous in Fable 5 is already available in other models on the market, including the GPT-5.5 by OpenAIearlier and more widespread than Mythos. However, that version of ChatGPT has not suffered any suspension. The asymmetry of treatment between OpenAI and Anthropic is evident. The geopolitical context is also important: the United States and China are immersed in a technological race unprecedented in recent decades and each power is playing its weapons, from tariffs to critical materials like rare earths to vetoes on chip sales, EDA software or the Export Control Reform Act 2018, where new technological categories fit, such as the most advanced AI models. Of course, for the Asian giant, every obstacle has worked as a kind of catalyst to advance faster and be increasingly independent of outside technologies. In detail. The order’s enforcement mechanism is supported by the “export control directive” managed by the Department of Commerce. In practice, it means that accessing Fable 5 or Mythos 5 as a foreign citizen (even if you are in the United States) is grounds for infringement. Anthropic tried to anticipate the problem with thousands of hours of testing with the US government, shared it with 40 organizations that manage critical infrastructure and then, with another 150 entities more precisely so that they could find and thus be able to correct vulnerabilities before a third party with malicious interests did so later. It hasn’t been enough. Because Mythos It is a specifically designed model for cybersecurity. Before its public deployment, Anthropic shared it with organizations that manage critical infrastructure precisely so that they could find and thus be able to correct vulnerabilities before a third party with malicious interests did so. Its blackout affects both commercial users and those entities in charge of active defense management. Yes, but. For now, Anthropic complies with the order, but makes it clear that it does not agree and that it is working to resolve it: “We believe this is a misunderstanding and we are working to restore access as soon as possible.” The company behind Claude says it supports the government in blocking really dangerous technologies, but that this process has to be transparent, fair and based on real technical facts. In short, he has not said the last word. On the other hand, the million-dollar question is whether the US government can apply the export control law, initially designed for chips, satellites or critical and specific software, on an AI model available on the internet and in general use. In fact, this regulatory movement puts on the table the importance of having a clear law that defines when, how and with what guarantees a government (in this case, the United States) can intervene. In Xataka | Spain does not want to wait for Mythos to wreak havoc on its companies: that is why it is going to ask for “early access” to Anthropic In Xataka | Claude Fable 5 is the most powerful public … Read more

Claude Fable 5 is the most powerful public AI model in history. Also the most expensive, exclusive and frustrating

by

When Anthropic presented Claude Mythos Preview two months ago, he did it with a singular message: it is so powerful that you will not be able to use it. That, of course, caused everyone to want access to it. Well: Anthropic has just introduce Claude Fable 5 and Claude Mythos 5its new AI models directly derived from that. There is good news, but also bad news. Like Mythos, but capped as a precaution. Anthropic already warned that Claude Mythos Preview was a spectacular tool for finding security vulnerabilities. That made it especially juicy for cybercriminals, so the company decided that only a few trusted entities (under its Project Glasswing) would have access to the model. That learning has now been applied, because in this announcement we have two different (and layered) versions of the model: Claude Fable 5: a model with all the capabilities of Mythos Preview, but with notable security measures that prevent it from being used for malicious purposes. As soon as the model detects that we are asking something “dangerous”, it avoids the question and even forces the use of an inferior model, Claude Opus 4.8. Clear examples: questions about cybersecurity or the development of biological weapons, for example. Claude Mythos 5: This version is somewhat less capable than Fable 5 in terms of cybersecurity, but will only be available to “a small group of cyber defenders and infrastructure providers.” It is the natural heir to Mythos Preview, and according to its creators it is even better than the original version. Claude Fable 5 / Mythos 5 simply sweeps the most demanding benchmarks on the planet. There have never been more powerful models. Anthropic’s internal testing shows that we are facing the most powerful AI models in history. In all benchmarks – including the new FrontierCode programming, much more demanding than SWE Bench Pro – the scores of Claude Fable 5 and Mythos 5 are simply spectacular, well above those of their rivals. The jump from Claude Opus 4.8 is really surprising, but it leaves GPT-5.5 and Gemini 3.1 Pro far behind (they don’t compare with the recent 3.5 Flash). This is a brutal blow to Anthropic’s table, and we will see how both OpenAI and Google respond. Claude Fable 5 is amazing. Ethan Mollick, well-known AI popularizer, has had access to Fable 5 for a few days and is amazed by the experience. With this model he has managed to complete projects such as east of the isochronic map that previous models had never solved, and it has done it almost “the first time”. In one of the cases Fable 5 worked for 9 and a half hours straight to produce a code called Concord of data analysis. Their conclusions are compelling: Last year (when working with GPT-5 Pro) I called him “work with a magician”: you recite the spell and something happens. With Fable, the spell has become so powerful that I’m no longer sure I’m the wizard. I feel more like a patron. I describe what I want, pay for it and evaluate the result. The conspiracy takes place somewhere I can’t see, in hundreds of small decisions over which I never have a say. Work has gone from being a process to being a result. I no longer direct; charge. The criticism is unanimous. Andrej Karpathy, who recently signed by Anthropic, commented on X how this is a qualitative leap that for him is of the same relevance as the one that Claude 4.5 represented in November. That model began the overtaking of OpenAI: this puts it even further away (at least, for now). Other tweetersemployees or not from Anthropic, make it clear that this is an important leap in the capabilities of AI models. It’s only been a few hours since the launch, but everything points because we are indeed facing a notable leap in quality. Consume tokens like there’s no tomorrow. But in the face of that fascination, the criticism. Discussions on Reddit reveal how users who have started using it have quickly detected the problems associated with this release. The first of them: Claude Fable 5 burns tokens like there is no tomorrow. Its consumption is enormous, and the quotas for Pro and even Max accounts run out in minutes if we use the model intensively. If it already seemed to us that we were exhausting the limits of the free or quick payment accounts, with Claude Fable 5 that feeling worsens: Fable 5 is fantastic, but we can barely use it often with the Pro or Max plans because those dreaded messages about waiting X hours to continue using it quickly appear. Extremely cautious. Anthropic has been very serious about avoiding misuse of Fable 5, and as soon as it detects anything suspicious it “brakes” and “downgrades” the model so that at that moment the one that is activated is Claude Opus 4.8 (which is not bad at all). The problem is that users are detecting that the model takes completely harmless prompts as dangerous. Although in Anthropic indicate Although these security measures are activated in less than 5% of sessions, what users are detecting is that they are activated much more. Fable 5 can get silly. Not only that: Fable 5’s own design means that if it encounters a prompt that it detects as dangerous, the model tries to avoid the response and automatically reduces your capabilities (‘nerfing’) without you knowing. It gets a little sillier on purpose, so to speak. As Anthropic itself explains on the system card, We have implemented new measures that limit Claude’s effectiveness in requests related to the development of cutting-edge large-scale language models (LLMs) (for example, in creating pre-training pipelines, distributed training infrastructure, or designing machine learning accelerators). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our security measures prevents giving an advantage to those users most willing to violate those terms. Unlike our cybersecurity, biology and chemistry interventions, and distillation attempts, … Read more

Comparing Apple’s AI to ChatGPT or Claude is a mistake. Apple is not playing that game

by

to whom They rule out Apple in the AI ​​raceeye. The company may have arrived late and it certainly may have little to show today, but its evolution over the last three years reveals three interesting things. The first, that Apple does have its own AI models. The second, that they are very far in performance from the best of OpenAI and Claude. Third, that may not matter at all. Three years of evolution. The trajectory of the technical documents shared by Apple in recent years reveals a series of more than relevant changes. In 2024 its initial proposal was limited to small models of about 3,000 million parameters (3B) specialized in solving basic tasks like generating Genmojis or text summaries. In 2025 the company launched its MLX framework to the developer community to facilitate the integration and use of local models. Now, in 2026, They propose a hybrid infrastructure based on a basic principle: Simple requests: they run in small local models on the device, you don’t even need an internet connection Complex requests: the system delegates the task to be processed in the cloud privately through Private Cloud Compute A (maybe) great idea: NAND can help. The most relevant milestone of Apple’s new approach lies in the design of its AFM 3 Core Advanced model. In today’s mobile phones we have a big bottleneck with the execution of capable (large) AI models because these devices have a very limited amount of memory (12 GB on some iPhones). To be able to fit a model with 20,000 million parameters (20B), Apple has decided to store that model in the internal SSD unit, not in memory. In the AFM 3 Core Advanced model the “experts” are in the mobile’s SSD. They are preselected and loaded into RAM to be used dynamically, optimizing model execution. Experts by prompt, not by token. It then activates a series of pruning techniques (Instruction-Following Pruning, or IFP) to activate only between 1,000 and 4,000 million parameters in a sparse manner (sparse), somewhat similar to what is done in models with Mixture-of-Experts architecture. But Apple selects these experts at the beginning of each prompt, not token by token, which allows it to avoid the slow bandwidth of the mobile’s NAND storage compared to its RAM memory. Privacy by flag. If for something Apple’s approach stood out from the beginning It was for his privacy.which is implicit when using local models. But if the request is complex, the system redirects it to the AI ​​models in Apple’s cloud, the Private Cloud Compute (PCC). Unlike other platforms and infrastructures such as those of OpenAI or Anthropic, conversations with Apple’s AI are encrypted and are totally private according to the company: this data is not shared with third parties (because not even Apple can see it) and it is not used to train its models. Five models with the help of Gemini. Although Apple is obsessed with total control of its products, this time had to give in and ally with Google so that their Gemini models could “show” Apple the way. The result is a third generation of models that are developed in collaboration with the Mountain View firm. We have five models in total: AFM 3 Core: 3B parameter dense model AFM 3 Core Advanced: sparse model of 20B parameters with activation of 1B to 4B parameters depending on the task AFM 3 Cloud: a powerful but also efficient and fast model that runs on the Apple cloud. ADM 3 Cloud (Image): for generating and editing images, the heart of both these options and the new Image Playground AFM 3 Cloud Pro– Apple’s most powerful cloud model is for autonomous agents. It has been trained with Google TPUs and runs on Nvidia GPUs within Google Cloud infrastructure Performance, an unknown. Unlike what other companies usually do when they present their models, Apple has not published metrics on known benchmarks. Instead, it shows “human preference” metrics in which it compares user satisfaction when using its models versus competing models. The comparisons are also with previous versions of these models, which does not clarify much what can be expected from them. But they are not in the race for the best model. In 2025 yes there was comparison with open weight models of that time (Qwen-3-4B locally, GPT-4o or Llama 4 Scout in the cloud) and then they seemed to be at a good level in reference to those options. Expect them to be behind the most recent models from OpenAI, Anthropic, or Google itself, and it’s unclear how they compare to the new Chinese open weights models. One thing seems clear: Apple is not very interested in having its own Mythos, at least for now. Your objective is different. Apple models from 2026 are “preferred” more than those from 2025. Logical, but also useless when it comes to understanding how good these models are compared to the competition. But integration is important. Apple’s big ace to compensate for this difference in capacity is that its models have full access to the user’s OS, apps and hardware. AFM models are integrated with iPhone camera sensors, notification history or local app permissions. This allows useful tasks to be carried out that an LLM that is “disconnected” from the hardware will hardly be able to replicate. Here the integration of the models with the hardware and software of the device is (or wants to be) fundamental. Beware of mediocrity. This approach focused on integration and privacy is especially striking and differentiating from its competitors, but there are risks. Among others, the product is limited by its functional capabilities compared to the competition. If local models do not solve and cloud models also do not behave reliably, Apple runs the risk of having an AI that is secure and private but technically mediocre in its responses. Siri has already been criticized for being especially stupid: Siri AI must precisely eradicate that perception. In Xataka | Apple has designed Siri AI so … Read more

copying what Claude Cowork does

by

One might think that the new Siri AI would be specially integrated on iPhones with iOS 27but in reality its options may end up being even more striking on macOS. Apple’s desktop operating system promises especially powerful integration because will be integrated into Spotlight: Once you start typing something in the Spotlight search field, the system itself will be in charge of detecting whether Siri AI should take action. And when it goes into action, things happen. To start, there will be a shortcut with the Ctrl key that will cause when you press it while clicking anywhere in the macOS interface, a contextual option will appear to make a call to Siri AI and that will give you information about that element on which you have clicked. But the really interesting thing is that in addition to those conversational searches and queries that one can also do on the iPhone, in macOS Siri AI can act on our files to, for example, extract information about them. How to enjoy Claude Cowork without having to pay for Claude Cowork During the presentation, one of those responsible for Siri AI at Apple selected three PDFs to compare the specifications and cost of three 3D printing products. Siri AI scanned the documents and generated a comparison with a comparison table in just a few seconds. This type of feature is especially interesting, because basically what Apple is offering here is its own Claude Cowork integrated into the operating system. An AI that can manage your computer for you—or at least, manage part of it for you if you let it—to assist you in all kinds of tasks. Typical examples of things you can do with Claude Cowork on a Mac include things like “organize my downloads folder by content type and move old .dmgs to the trash”, “go to my supplier’s website, download the invoices for the last month, extract the amounts and fill out this internal form”, or a simple “My desktop is full of files and folders, organize it and leave it clean.” It seems that these types of options are now coming in that Siri AI integration in macOS, and Plus they do it for freetaking advantage of the alliance between Apple and Google and those new local models that will be installed in the operating system to resolve these types of queries. There’s another interesting bonus here: Apple has promised that all of these AI features are built with privacy as a key focus. In theory, this will mean that nothing you do with these types of functions will be recorded on any company’s servers. Certainly not in Apple’s, which will use your Private Cloud Compute to guarantee precisely that if local models are not enough, when we have to use the cloud models that Apple offers us, those queries will remain private. If that integration is as powerful as it promises, we may have here a real punch at the table of Apple and Googlewhose alliance will make maybe (just maybe) Claude subscribers realize that they don’t need to pay if macOS already offers basically the same thing for free. We will see if that promise is fulfilled when we can test this operating system. In Xataka | The alliance between Apple and OpenAI has begun to crack: there is an underlying reason that explains it, according to Bloomberg

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.