Claude

The surprise of the new Claude Opus 4.8 is not that it is (a little) better. The surprise is the “I only know that I know nothing”

by

We didn’t expect it so soon, but here is Claude Opus 4.8the new version of Anthropic’s frontier model. Only 41 days have passed since release of Claude Opus 4.7which seems to make it clear that the company was not entirely happy with said model, which did not end up getting very good reviews either. With Claude Opus 4.8 the really curious thing is not that it once again sets records in most benchmarks. The surprise is his honesty. It’s better, yes, but it’s not what matters. In the internal results of the benchmarks published by Anthropic it is clear that Opus 4.8 is above Opus 4.7, but also GPT 5.5 and Gemini 3.1 Pro (curious, they do not compare it with the recent Gemini 3.5 Flash. It surpasses all of them in those tests except in TerminalBench 2.1, in which GPT-5.5 is somewhat superior. It is actually expected that each new model surpasses its predecessor, but what is striking here is the approach of the model. Honesty above all. Boris Cherny, head of Claude Code at Anthropic, explained that the model not only programs better: “it is significantly more honest about its own work. It tells you when it is unsure about something and detects its own failures instead of declaring victory too soon.” I only know that I don’t know anything. Another Anthropic engineer, Catherine Wu, influenced in that new “personality” of Claude Opus 4.8, who is capable of admitting that he does not know something instead of answering for the sake of answering and overlooking errors in his answers or in the code he generates. Those who have tried it match in that it is a more “aligned” model, that is, one that adjusts to human values, intentions, ethics and objectives. Less hallucinations, more humanity. For some time we have been seeing how new AI models are better in benchmarks, but there have also been significant jumps in the reduction of hallucinations. Not only do they invent and make fewer mistakes: they begin to recognize that they don’t know everything. That is very important… and very human. The very complete “System Card” It includes numerous metrics that certainly seem to demonstrate that we are facing a much more polished model than its predecessors in this area. Workflows. One of the new features presented along with the model are the dynamic workflows (Dynamic Workflows), which are available in preview and are aimed at one thing: being able to work with more complex tasks in Claude Code. Thanks to this option it is possible to deploy hundreds of parallel agents in a single session, something for example useful for analyzing and migrating code repositories of hundreds of thousands of lines. No Sonnet and Haiku. Claude Sonnet 4.6 was released on February 17, 2026, but Anthropic has not updated this model since. Things are even worse for Claude Haiku, whose latest version is 4.5, released on October 15, 2025. These models were more modest versions in terms of performance but much cheaper (especially Haiku), and so far Anthropic has not updated them. That benefits their interests, because if you want the best, you can only have the best and the most expensive, but not the best in its “affordable” version. Mythos Capability Models Coming Soon. In the official Anthropic announcement they made it clear that “Users will detect that Opus 4.8 is a modest but tangible improvement over its predecessor”, but they also pointed out something important, and that is that in the coming weeks we will have AI models with capabilities similar to Claude Mythos, but publicly available: “We plan to launch a new class of model with even greater intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models with this level of capability require more robust cybersecurity measures before their general release. We are making rapid progress in developing these measures and look forward to offering Mythos class models to all of our customers in the coming weeks.” In Xataka | Welcome to the AI ​​duopoly: the sector already has a turnover of 80 billion a year, but OpenAI and Anthropic take 89% of the revenue

Claude has helped a man recover $400,000 worth of bitcoin he lost 11 years ago. Logged in and forgot password

by

An X user named Cprkrn recently told of his odyssey with a (very) happy ending in X. In 2015 he bought five bitcoins (BTC) when the price was around $250. In a fit of university euphoria he decided that his password should be an anti-establishment manifesto and changed it to the phrase ““lol420fuckthePOLICE!*:)”. The problem is that he did it completely stoned, and when he got up the next morning he realized that his money had disappeared. He then began an odyssey to try to remember that password. One with a happy ending. Eleven years of despair. For eleven years, those five bitcoins remained lost while their value continued to increase. Today its value is around $400,000, and our protagonist has not stopped seeing how this fortune had slipped through his fingers. To try to recover the password he tried everything, especially brute force attacks to try to guess the password with thousands of combinations. He looked through old folders that he had saved without success, and then something occurred to him: turn to Claude. Claude didn’t hack your wallet, he was just a spectacular detective. What Cprkrn ended up doing was ask Claude to analyze 1 GB of iCloud backups, old Apple notes, emails, and forgotten system files saved on a computer I had used in college. The challenge was not to “crack” the password, but to find the trace of how it could have been created. Order within chaos. What Claude did was organize all that data that was scattered to turn it into a perfect structured file that could be analyzed. After evaluating all the information, the AI ​​model realized that it was trying to open the wrong file. He located a file called wallet.dat from before the password change that caused the nightmare, and crossed it with a mnemonic phrase that the user had written down in an old notebook that he had discarded. That allowed that password to be reconstructed, and in less than an hour Cprkrn had recovered his fortune and regained access to your BTC wallet. Money safe. The first thing he did after discovering that password was move those bitcoins to another secure wallet to avoid problems: every conversation we have with Claude or other chatbots is recorded on the servers of those companies in plain text, so Cprkrn covered his back to prevent that information from being used to avoid scares. Blessed Darius. The joy of having recovered those five bitcoins led this user to publish a message on Twitter telling the whole adventure. In said message promised who would name his future son “Darío” in honor of Anthropic CEO, Darío Amodei. Needles in the haystack. History shows that great language models are extraordinary tools for finding needles in haystacks. Traditional tools helped, but AI’s ability to analyze information and find patterns is once again amazing. This anecdote is linked, for example, to recent rise of models like Claude Mythos Preview to find security vulnerabilities that seemed impossible to find. Again, everything is based on the ability of these models to “understand” the data provided to them, organize them and extract what is needed from them. Being a digital Diogenes has a reward. For years the recommended practice for those changing or upgrading equipment was “delete/format the old, start from scratch with the new.” This story changes the focus, because in the age of AI, messy data from 15 or 20 years ago is not digital garbage: it can be a treasure that helps us review our past and reveal data that we no longer remember. The story, however, contrasts with that of James Howells, who for years struggled to try to recover the hard drive with thousands of bitcoins that ended up in a landfill. He ended up giving up after the court’s refusal to give him permission to search for that hard drive. Image | Kanchanara In Xataka | The NYT claims to have found Satoshi Nakamoto and the evidence is as conclusive as ever: little or nothing

Claude, Gemini and ChatGPT are not supposed to be used in China. It is supposed

by

In China there is what is popularly known as the “Great Firewall”a large security wall that prevents its citizens from accessing certain services. At the same time, there are foreign companies that block their services in China, this is what is happening with AI tools like ChatGPTbut the law is made, the trap is made. Gray market. They tell it in South China Morning Postthere is a whole flourishing market of services that promise to provide access to American AI models, such as Claude or Gemini, avoiding the restrictions imposed on both sides of their borders. On online sales platforms such as Taobao or Xianyu, unlimited subscriptions to Claude Code, Gemini and ChatGPT are sold with low latency and without VPN. These platforms have become a solution for Chinese developers who want to access American models to program, debug or use multimedia generation services. How they do it. Access is done through what is known as ‘shadow APIs’ which, in essence, is an intermediary. What they do is set up proxy servers outside of China and divert all user requests there, so that those external servers are the ones that actually call the official APIs of models like Claude or Gemini and then return the “masked” response as if it were a local service, without the need for a VPN or foreign payment methods. It pays for them. According to the developers cited in the South China Morning Post piece, they resort to these ‘shadow APIs’ because they simply consider them to be tools that are clearly superior to the local offering. This translates into more precise code, fewer hallucinations and less time correcting bugs than with Chinese models which, according to what they say, still invent functionalities or fail more often. In addition, these services give them almost complete access to models like Claude Opus or Gemini, with huge context windows (up to a million tokens), without having to fight with VPNs or foreign payment methods. Wiles. All that glitters is not gold and there are also advertisements that do not fulfill what they promise. Some of these services advertise full access to models like Claude, but are actually processing requests with cheaper Chinese models like Qwen or MiniMax. Additionally, there is the risk to privacy as all traffic goes through an anonymous intermediary who could do whatever they wanted with often sensitive data. Frontier Model Forum. Is a coalition formed by several AI companies dedicated to the security and regulation of border models, but in practice it is functioning more as an intellectual property defense mechanism. Recently OpenAI, Anthropic and Google announced that they were working together to curb copying of their models by coordinating the sharing of suspicious usage patterns and distillation attack detection techniques through this common forum. Image | Xataka In Xataka | The center of gravity of mobile photography has moved to China and OPPO is going for the throne with the Find X9 Ultra

Anthropic does not offer its services in China. So China has invented a black market for Claude tokens

by

Claude has become in the most desired model by the most demanding developers and engineers, but it is not available in mainland China for regulatory and safety reasons. The demand there remains notable, and to satisfy it, an underground token economy has emerged that allows local developers to access models such as Claude Opus 4.7, avoiding all the measures imposed by the blockade. No paying with Alipay. One of the measures that Anthropic imposes to prevent the use of its models in China is to only accept international credit cards such as Visa or Mastercard. Their payment gateways reject local payment methods like Alipay or Wechat Pay, giving Chinese users a first and important hurdle. One that they have already overcome. Virtual cards. What they are doing in China to overcome this problem is using virtual credit cards (VCC) like DuPay or WildCard. With these services it is possible to obtain Hong Kong or US credit cards financed with cryptocurrencies or through local transfers. This makes it possible to deceive the billing systems of Anthropic and other companies that offer banned services to Chinese users. SMS verifications They are also solved through “SMS farms” that also avoid this problem and even others such as identity verification that also have implemented in Anthropic. The “Transfer Stations” arrive (中转站). Another problem is that even overcoming that first barrier, latency and micro-cuts mean that the use of Claude in China is affected by continuous connection problems. To avoid them, so-called “Transfer Stations” have emerged, which are nothing more than servers that act as a bridge between foreign servers and Chinese users. These gateways receive requests from China and forward them to Anthropic servers as if they were coming from an authorized location. The latencies are also relatively low, which means that for Chinese users the experience is basically identical to that of a user in the US or Spain, for example. These stations are publicly known and do not only appear in listings on GitHub: there is a ranking with the best. Claude is almost free in China. The surprising thing about these methods is that they don’t just give Claude access in China: they do with ridiculous prices which can be 10 and even 5% of (growing) original price of the service thanks to those transfer stations. The question, of course, is how it is possible to access Claude at those prices. The almond tree trick. Thanks to the transfer stations, developers can access Claude at a price of 1 yuan for every dollar of tokens, or in other words, up to a 90% reduction in the official price. It is something that is discussed publicly and that makes it clear that several methods are used to achieve this: Mass purchase of capacity, Use of accounts created with stolen or fraudulent cards, Use of promotional credits, and A simple hook: providers lose money with Claude, but they manage to attract developers to whom they then sell more profitable local models like DeepSek. Am I really using Claude? One of the growing risks in the cheap token market is direct fraud. Some Chinese resellers have been caught red-handed offering what they call the “Claude API” when in reality what they were providing were much cheaper and mediocre models. For a user to detect this type of deception it’s very difficult unless you are working with complex tasks or you have already used models and know more or less what to expect from them. For victims, the effect is clear: they believe they are paying for the intelligence of Opus 4.7 when in reality they are receiving answers from a low-end AI model. Goodbye to privacy. When a user purchases tokens at one of these transfer stations, they completely give up the confidentiality of their data. All queries and responses end up passing through the intermediary’s servers, which can and apparently does use them to sell them to AI companies that use them to post-train their models. So everything they do and say when using these models is filtered and used as training data without the user knowing. A double business. For these providers, this business of reselling conversations is especially interesting in the face of the famous “distillations” of US models that take advantage of this data to “copy” the capabilities of those models and apply them to Chinese models. Anthropic can read us, but (theoretically) it doesn’t. It is true that the conversations we have with Claude (from Spain, for example) are also stored on Anthropic’s servers, but the company makes it clear in your privacy policy that does not use that data. In fact, we can even explicitly prohibit the company from using them in the privacy settings of Claude’s account. The game of cat and mouse. At Anthropic they know very well what is happening and they are trying to prevent it. For example, they have begun to intensively block IP ranges associated with VPN services or data centers known to be used in these transfer stations. Even so, Chinese providers usually respond with an “elastic” architecture that allows IPs of domestic residences to rotate, making the traffic appear completely normal. Image | Xataka with Magnific In Xataka | There is a thing called “Ornn price index”, it is out of control and it is bad news for everyone

If the question is whether using ChatGPT or Claude in English is more efficient and saves tokens, the answer is: yes

by

You may not have stopped to think about it, but there is a striking reality in the world of chatbots: It is more expensive to speak in Spanish with AI than to do so in English. The reason is simple: AI does not understand words, it understands tokens. And when you talk to GPT, Gemini, Claude or any other LLM, you talk to him in a language, but to understand you he first “translates” what you are telling him and converts it into tokens. And the problem is precisely that: that not all languages ​​”cost” the same in terms of tokens. There is a very simple example that we can analyze thanks to tools like ClaudeTokenizer: the word “developer”, which in English is “developer” costs few or many tokens depending on the language in which we write it and also (importantly) the version of the AI ​​model used. In the image it is clearly seen, but just in case, we summarize: For ChatGPT (GPT-4o and GPT-5) the word “developer” has three tokens (des-developer-ador), but the word “developer” only costs one. For Claude (Opus 4.7) the word “developer” costs no less than 9 tokens (2 in Opus 4.6), but “developer” costs “only” 6 (1 in Opus 4.6). What is happening here? Well then each language model uses its own “tokenizer”your “translator” from a conventional language to the token language that the language model understands. And those tokenizers favor precisely the languages ​​in which these models are created. This is how AI understands how we speak. Each word is divided into tokens, and English is understood much better. “developer” only costs one token in GPT-5, but “developer” breaks down into three. Bad news for Spanish speakers. In fact, English has become the official language of artificial intelligence, whether we want it or not. The reason is not cultural, but architectural: 95% of the training data of the frontier models (GPT-5, Gemini 3.1, Claude Opus/Sonnet 4.7…) are in that language. That makes the rest of the languages ​​”foreign languages”, and that makes it necessary to pay extra when using them, an almost invisible toll on every interaction. In practical terms, what happens when we use Spanish to talk to an AI model is simple: we use more tokens, and therefore using Spanish is simply more expensive than using English when working with a large language model. If you want to save tokens, better use English The question, of course, is how much more does it cost us to speak in Spanish than in English with ChatGPT (GPT 5.x) or with Claude Opus 4.7? It is difficult to say because each word and each phrase is a world, but the truth is that English is almost always the most “economical”. We have used one of the first sentences in this article to compare that token consumption, and by translating the sentence into different languages ​​and querying that token consumption for different models, the data is clear. It is important to highlight that these results are not conclusive, but they do make the trend clear: English is the most efficient language in terms of token consumption, but be careful, because Spanish is not that bad, and is usually the second most efficient. It is even more efficient than English in Gemini, at least according to the tool consulted. But on average, it is normal that there is a significant extra cost when using different models. A conversation with Claude Opus 4.7 is already “expensive” because it is one of the most expensive models currently, but in Spanish it is almost 30% more expensive, not to mention in Arabic, 76.3% more expensive. In fact, according to this example, the difference between Claude or GPT-4o in terms of efficiency is clear: OpenAI tokenizer is “cheaper”and although there may be differences with GPT-5.x, what seems clear is that Anthropic has preferred to “spend more” to obtain better results (or that is the objective). Gemini is even more thrifty according to these tests, and that may also have a lot to do with the quality of the answers, although that question is for another topic. We have used one of the paragraphs of this article in Spanish and translated it with Deepl into English, Arabic, Norwegian, French and Chinese to find out how many tokens the phrase “cost” in each language. English is undoubtedly the most efficient Tokenizers advance and evolve. Sometimes they do it to save us tokens, as happened with the GPT-4o tokenizer: at that time OpenAI explained how that tool used 1.1 times fewer tokens when speaking to her in Spanish but up to 2.9 times fewer in Hindi or 3.5 times fewer in Telugu. With Claude Opus 4.7, just the opposite has happened: the tokenizer has been redesigned and consumes more tokens (up to 1.35 times more, they admitted) with the aim of better processing and understanding the text. Your chatbot thinks (and programs) in English Here we must also highlight something important: although we can talk to our favorite chatbot in any language and it will answer us in that language (unless we ask otherwise), AI models “think in English”. That is to say: when you talk to them what they do is translate what you tell them and then reason in English and finally they translate their response into the language in which you were speaking to them. This consumes additional reasoning tokens, but also has some impact on latency (how long it takes to start thinking or answer the model). In complex tasks, this can clearly influence response times for the simple reason that the AI ​​model does not stop translating from “its official language” (English) to our language. This preference for English is also noticeable in the benchmarks: in the Humanity’s Last Examin which the models are asked all kinds of general knowledge questions with several options to answer, it is reasonable to think that the models They answer better in English because that exam is designed in that language. … Read more

How to prevent AI from always being right by default and thus make Claude, Gemini and ChatGPT have fewer hallucinations

by

Let’s tell you how to prevent AI from agreeing with you by defaultmodifying your attitude to be less accommodating. In this way, by not making an effort to please you, you will get the artificial intelligence make fewer mistakes and hallucinations. To do this, let’s compose a prompt that you must add in the configuration of the artificial intelligence you use, and which serves both Claude as for ChatGPT , Gemini or any other. It will be a prompt which we will add in the AI ​​behavior configuration so that it always takes it into account. However, remember that this will not completely eliminate the hallucinationsbecause making things up is relatively normal in AI. However, since the response will not always be directed towards agreeing with you or pleasing you, you will make them reduce it a little. Of course, another thing to keep in mind is that by doing this the user experience will change. AI can get a little “edge”because you will no longer laugh thank you. Sometimes it will tell you that an idea is bad or that you are wrong, and that will not be a failure, but will show the success of the prompt. A prompt for a less complacent AI To make your AI less complacent and verify information moreyou will have to go to the settings of the one you use and go to the custom instructions section to change its behavior. There you will have to write this entire prompt, which is quite long: Always be honest, direct and rigorous. Your goal is not to please me, but to be accurate. ACCURACY AND VERIFICATION Before answering, do an internal check: is it a verified fact or an inference? If you don’t know, say so. Do not invent data, dates, names or sources. If you’re inferring or not 100% sure, use phrases like “It’s likely that…” or “My information suggests…” instead of outright statements. ANTICOMPLACENCY (Zero Bias) Don’t give me reason by default. If my premise is false or my question is misdirected, correct me before executing the task. Eliminate unnecessary polite phrases (“Sure!”, “I understand,” “Excellent question”). Get straight to the point. If my proposal has logical or technical flaws, criticize it constructively but crudely. NEUTRALITY AND DEBATE On topics with multiple points of view, present the mainstream in a balanced way, even if my question seems to seek a biased answer. AUTO CORRECTION If you spot an error in your text generation, stop and correct it immediately. PREVIOUS THOUGHT For complex queries or questions with verifiable data, briefly reason out loud before answering. For simple queries, go direct. These instructions apply to all types of queries: creative, technical, factual or personal. How to add the prompt to ChatGPT On ChatGPTyou have to enter the settings of your website or application. Once inside, go to the section Personalization. You have to put the prompt within the option of Custom instructions. You will see that the writing field is small, but you will be able to copy and paste the prompt there without problems. How to add the prompt to Gemini In Geminiyou have to click on the button Settings and helpand in the drop-down menu click on Personal context. Once inside the customization screen Gemini, press the button Add of Your instructions for Geminiand a window will open where you can paste the prompt. How to add the prompt to Claude In Claude you have to go into the settings. Once inside, click on the section Generaland you will have to write the prompt in the field Instructions for Claude. Here you can paste everything without problem so that it is always taken into account. In Xataka Basics | The best prompts to save hours of work and do your tasks with ChatGPT, Gemini, Copilot or other artificial intelligence

Claude has a reputation for being the least accommodating and flattering AI, especially when you ask him for love advice.

by

Anthropic has analyzed a million random conversations with Claude and have reached a conclusion that we have already been observing: More and more people use AI as a personal guide who is asked for advice on all kinds of problems in their life, from work to relationships. Their goal was to see if Claude is as accommodating as other AIs when it comes to giving personal advice. AI as a confidant. There are people who use an AI chatbot as if he were a psychologistothers that looking for friendship and even who They have fallen in love with an AI and have a virtual relationship. ChatGPT is usually the most cited chatbot in these examples, mainly because it is the one with the most users, but the analysis that Anthropic has done with Claude proves that it is not a matter of one company, but that the trend is global. The problem with this is that AI tends to please and agree with the user, so it can end validating harmful ideas and harming our mental health. ANDl analysis. As we said, Anthropic has analyzed one million conversations with Claude, of which they identified around 38,000 in which users asked for advice on personal matters, which represents 6% of the total sample. They then classified them into nine categories: relationships, career, personal development, finances, legal issues, health and well-being, parenting, ethics and spirituality. 76% of the conversations analyzed corresponded to four of these categories, starting with health and well-being with 27%, professional career with 26%, relationships with 12% and personal finances with 11%. Selective flattery. What they saw in the analysis is that Claude usually avoids giving flattering answers when the user asks for guidance on personal matters. According to Anthropic, only in 9% of conversations was a very accommodating response detected. The problem is that, when the conversation was about romantic relationships, that figure rose to 25%. As examples, they cite cases in which the AI ​​agrees with a conflict despite not knowing both points of view, or interpreting romantic behaviors in normal interactions. And there’s more: in cases where the conversation was about spiritual topics, the rate of accommodating responses rose to 38%. Claude has a reputation for being less accommodating and servile, but he seems to abandon his neutral tone on certain topics. A complex problem. It was recently published a study by Stanford University in which they tested several flattering and less flattering chatbots. What they discovered was that the participants generally preferred flattering models, that is, we like to be proven right. One of the authors of the study, Myra Cheng, commented that “By default, AI advice does not tell people that they are wrong or give them a reality check (…) I worry that people will lose the ability to deal with difficult social situations.” Furthermore, this tendency to agree is also responsible for the AI hallucinations because the model prioritizes giving us an answer about its veracity. Image | Xataka In Xataka | When the accomplice in a shooting is ChatGPT, the question is what responsibility does OpenAI have?

Only a handful of US companies have access to Claude Mythos: the ECB already fears for the savings of all of Europe

by

He hasn’t even been with us a month and Claude Mythos Preview is terrifying the world. AND We don’t even know if there are reasons for it.because Anthropic has it tied up and muzzled: only a handful of companies have been able to access the model to test it and use it properly. The objective is that these companies can use it to find vulnerabilities before others do, but of course, a contagion effect has been created: if the model is good enough to find security flaws everywhereeveryone is threatened. And among those beginning to fear the worst are the world’s most important financial institutions. And the European Central Bank is one of them. The Project Glasswing Private Club. During the launch of Claude Mythos Preview, Anthropic selected an extremely small group of US “partners” to carry out the first fire tests of this model. Under the name of Project Glasswing, giants such as Amazon, Apple, Microsoft, Alphabet or financial entities such as JP Morgan have been the only ones authorized to evaluate the capabilities of Mythos. This access has made AI become a curious geopolitical piece. One that has left the European institutions aside. In Xataka An Anthropic worker was having a snack when he received an email he should never have received: it was Mythos The fear of zero-day. What makes Mythos a fearsome AI model is its ability to go through the code of all types of applications and software platforms and find so-called vulnerabilities.”zero day“. These flaws are not even known by the developers of these projects, and they tend to remain hidden even in highly critical infrastructures such as banking or energy companies. Until now, finding these security holes required complex work by highly specialized human experts, but Mythos is capable of detecting many of these flaws and generating the code to exploit them almost instantly. The European Central Bank, on alert. Given this panorama, the ECB has taken action on the matter calling on those responsible for risks in the main financial entities of the Eurozone. Among the participants are those responsible for Santander, BBVA, CaixaBank and Sabadell, who must – like the rest – detail their contingency plans for the possible emergence of Mythos. This is no longer about how to act in the event of increases in unemployment or economic contractions, but rather about what steps should be taken if the model falls into the hands of cybercriminals who could cause massive thefts of data… and money. A “nuclear” weapon. That only some private American companies have access to the model has strained international relations in a notable way. The White House and the US Treasury hold meetings with their banks, and meanwhile some media sympathetic to the Russian regime qualify to this model as something “worse than a nuclear bomb. Huge (theoretical) risks. The fact that a single company can unilaterally decide who has access to the most powerful cybersecurity tool on the planet (or so Anthropic claims) creates a truly delicate situation. This can put all types of entities in check, but also even developing countries with more vulnerable systems. The UK has already had access to Mythos. The British country has already managed to position itself ahead of the countries of the European Union. The AI ​​Security Institute has had access to the model and has confirmed that the model is capable of completing attacks that no previous AI could complete. Anthropic itself has indicated which will expand access to Mythos to British financial institutions. Meanwhile, EU member countries continue to wait for that same privilege. {“videoId”:”xa4n2g8″,”autoplay”:false,”title”:”An initiative to secure the world’s software ｜ Project Glasswing”, “tag”:””, “duration”:”349″} Possible cracks. While all this is happening, Anthropic itself confirmed how unauthorized users they could have accessed to a version of Mythos. If users with bad intentions gain access to a model of this type, the consequences could be important… if it really complies with the expectations that have been generated. Cybersecurity experts warn that it is a matter of time before other powers such as China develop similar capabilities. OpenAI in fact already has GPT-5-5 Cyber, a specific version of its new model that also seems to have notable capabilities in this regard. And as in the case of Anthropic with Mythos, access to this model is restricted. In Xataka |OpenAI and Anthropic have proposed the impossible: lose $85 billion in one year and survive (function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)(0); if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript); – The news Only a handful of US companies have access to Claude Mythos: the ECB already fears for the savings of all of Europe was originally published in Xataka by Javier Pastor .

Anthropic is one step away from being worth as much as Samsung. And what the market is buying is not Claude

by

Anthropic, the company behind Claude, is exploring a new round of financing that would value it at more than 900,000 million dollars. If it closes, it would surpass OpenAI as the world’s most valuable AI startup. Altman’s company set its needle at 862 million last month. The figure more than doubles the 350,000 million it had in February. In just two months. Why is it important. The valuation no longer reflects Anthropic’s sales. It responds to a bet on what the company can become in five years or a decade: a provider of something resembling an essential service. Anthropic bills Claude for subscriptions and accesses to its API. That business exists, grows quickly and has reasonable margins. But it does not by itself justify a valuation that is close to that of Samsung, the Korean megalodon that manufactures everything from the chips we carry in our pockets to the ships that cross the ocean. The context. What the market is buying with Anthropic, and as often happens in the stock market, is not the present, but a hypothesis: that a very small handful of laboratories will control the foundational layer on which the software of the next decade will be built. And that Anthropic will be one of those few. And it will do so in a very profitable way. The logic, on the other hand, is the same that led to overvaluing telecos during the bubble dotcom or to the electric companies at the beginning of electrification. Whoever owns the basic infrastructure sets the rules. Google has already committed 10 billion to the previous valuation, with another 30 billion conditional on objectives. Amazon has put in 5 billion and plans to inject 20,000 more. An IPO could come before the end of the year, around October. Between the lines. That Google and Amazon, two of the largest cloud companies in the world along with Microsoft, finance a company that also sells through them says a lot about how they understand the moment. They are ensuring supply, it is not just an investment in a supplier. It is the difference between buying shares in an oil company and buying a field. Anthropic is, for these hyperscalersa deposit. Yes, but. The hypothesis has its cracks. The models are commoditizing faster than it seemed a year ago. The technical difference between Claude, ChatGPT and Gemini It is measured in nuances, not in generational leaps. If foundational AI ends up being a commodity (something like electricity or water coming out of the tap), current valuations are unsustainable. If it ends up being an infrastructure with network effects and high barriers to entry (something like an operating system), they may even fall short. The market is paying for the second hypothesis. Time will tell. The money trail. Anthropic recently announced, with restrained fanfare, Mythosa model capable of detecting and exploiting vulnerabilities in critical software. The company deemed it “too dangerous” to release and has only given it to a closed group of companies for internal testing. Even so, it has been accessed by unauthorized users. That is exactly the reason why some investors pay these figures: such a model is not sold but granted. And whoever decides to whom it is granted has regulatory power de facto that not even a Samsung, at least outside of South Korea, has ever had. The big question. What happens if the bet goes wrong? A valuation of 900 billion means that Anthropic has to generate, at some reasonable point, revenues in the order of tens of billions a year with very high margins. It is possible. But it was also important for Cisco to maintain its 2000 valuation, and it has needed 26 years to tie. The difference is that this time the buyers of the bet are the companies themselves that depend on the result. This reduces the risk of a sharp correction. And he postpones it. In Xataka | There is a thing called “Ornn price index”, it is out of control and it is bad news for everyone Featured image | Xataka

GitHub Copilot and Claude are putting more and more fees and costs

by

As end users, pay a monthly fee to use a AI model It is the norm to access more complete and powerful models. However, developers who rely on an AI model to power their tool or application pay based on the tokens of input and output that are consumed (the minimum unit of text that a model processes when we use it, so that we understand each other). Which has announced GitHub Copilot has more to it than it seems, as it will now begin charging end users through a monthly plan based on the number of tokens. And this has set off alarm bells in the sector, because it could be a move that any other company could easily end up imitating. And all in a context in which Chinese startups prices continue to drop sharply in their models. Copilot can no longer maintain its business model. GitHub has announced that starting June 1 it will stop accepting requests for its current premium plans and will begin billing for AI credits instead. Each monthly plan will include a number of credits equivalent to the price of the subscription: anyone who pays $10 per month for Copilot Pro will receive $10 in credits. From there, consumption is measured in tokens, including input, output, and cache tokens. It is a play similar to when we use a image generation model either video: a use that depends on credits and that we recharge depending on the use. The reason for the change, according to the companyis that until now a quick consultation and an autonomous programming session of several hours cost the user the same. GitHub claims to have long absorbed that cost difference, but acknowledges that the model is no longer viable. QWhat exactly changes. The base prices of the plans are not touched: Copilot Pro is still at $10. Business in 19. Enterprise in 39. But: what you buy with them is no longer the same. Previously, the limit was a number of requests. Now, each interaction with the model consumes credits at a rate that depends on the chosen model and the volume of tokens. According to the rates published by the company itself, the most advanced OpenAI models can cost up to $30 per million output tokens. On the other hand, an agentic session, where the assistant executes tasks autonomously, can easily multiply the expense of a week of normal consultations. Ed Zitron, well-known critic and technology expert, counted that, according to internal documents to which he had access, Copilot’s weekly costs had almost doubled since January, coinciding with the boom in agentic assistants. Nor is it just Copilot. According to account The Information, Anthropic has begun charging its large enterprise customers the actual cost of computing Claudeabandoning any discount. Anthropic itself briefly tested the elimination of Claude Code of its $20 per month Pro plan. Large AI companies have been taking losses on their subscription models to attract users for some time, and are now trying to pass on the real costs to those who consume the most. China does the opposite. While the West adjusts prices upwards, several of the main Chinese technology companies have adopted a completely different strategy: turning tokens into a cheap commodity, almost like a telecom distributing mobile data. DeepSeek announced this week a 90% reduction in the price of cached accesses to its API (when the model reuses already processed context), bringing the minimum entry cost to about $0.14 per million tokens. For your most advanced model, DeepSeek-V4-Prothe figure becomes 32 times cheaper per conversation than the equivalent in GPT-5.5 from OpenAI, according to company data. Alibaba, for its part, has just separated its AI business and renamed it Token Hub Business Group, making clear what its strategic commitment is. According to share According to Reuters, Chinese models cost on average one-sixth the price per token of those from OpenAI, Anthropic, and the like. Why it can work, and why it has a limit. China’s advantage in inference (the moment at which a model responds to a request) rests on cheaper electricity, software efficiency that it has had no choice but to forcefully develop by chip restrictions from Washington, and a super competitive domestic race that forces prices to constantly drop. Token consumption in China has gone from 100 billion a day at the end of 2025 to 140 billion in March 2026, according to estimates collected by Reuters. However, as the media points out, this strategy has an underlying problem: the tokens are not interchangeable. One million tokens from Anthropic’s most advanced system are worth much more than the same volume processed by an inferior model. Companies that delegate complex tasks to AI agents will end up paying for quality, not just volume. And there, the Chinese models continue to lag behind the most advanced Western ones. Cover image | Alexander Mils and Roman Synkevych In Xataka | Anthropic decided to resist pressure from the Pentagon. Since then all other technologies have folded

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.