Good

There is a battle to have the AI ​​model that programs best. And a good, pretty and very cheap rival has appeared in it: Cursor

by

Cursor has introduced Composer 2.5a generative AI model specifically intended for one thing: programming well. How good? Well, according to this startup, it does it as well as the best models of the moment, Claude Opus 4.7 and GPT 5.5, but it also does it for a lower cost. The challenge is striking not only because of what it means for Cursor, but because of how they have created that model: it turns out that it is based on a Chinese AI model. AI models specialized in one thing. While OpenAI and Anthropic try to develop general-purpose models—they do a lot of things really well— Cursor you have decided to focus on a specific task. The AI ​​startup has created an AI model specialized in programming, and has done so by arguing that a billion parameters are not necessary to compete with the best. Devoting yourself to a single thing allows you to not only gain efficiency, but also costs. This is not a decathlete, but a specialist in the 200 m event, so to speak. As good as GPT-5.5 or Claude Opus 4.7? That’s what they say in Cursor, because according to their tests with several specific programming benchmarks, the performance is on par with those two models that today are the great references both in programming and in other areas. And much cheaper. These results are also especially interesting when we add the cost factor. The average cost per task in the CursorBench 3.1 benchmark showed that Composer 2.5 managed to solve almost 65% of all tests for a cost of just $0.3. Opus 4.7 max and GPT-5.5 xhigh managed to reach that 65%, but at much higher costs: just over 4 dollars in the case of GPT, and 11 dollars in the case of Opus. The difference is abysmal. He API access price demonstrates the differences: 0.5 dollars per million input tokens 2.5 dollars per million output tokens, when Claude Opus 4.7 is 5/25 and that of GPT-5.5 is 5/30 respectively. Textual feedback. Unlike models that only learn from the final result, Composer 2.5 has been trained with a reinforcement learning technique (Reinforcement Learning) that allows us to offer clues about what is happening if errors are being made. This allows the model to recalibrate and act as a transparent teacher. One that also corrects word by word as it solves the exercise, not just when seeing the final result. 85% of the training budget has been dedicated exclusively to reinforcement learning, calibrating the model not for chat, but to execute code refactorings or fix bugs in real time. A model “born” in China. Those responsible for Cursor themselves have explained that Composer 2.5—like its predecessor, Composer 2launched at the end of March—is a model derived from Kimi K2.5, the AI ​​model of the Chinese startup Moonshot. Although that is the basis, already in Composer 2 the training and post-training tasks manage to improve the behavior in a very notable way in programming benchmarks and also in others such as Terminal Bench that evaluate the agentic behavior of these models. Cursor gets older. This startup became famous for creating a programming AI agent that was a pioneer in that fever we live for vibecoding. The user experience is no longer that of programming, as in traditional IDEs (Integrated Development Environments), but rather that of directing the machine to program it for you. Composer 2.5 doesn’t just program: it understands the structure and relationships between files, and turns Cursor into a much more competitive AI company, because it no longer depends on being able to work with Anthropic or OpenAI models, for example. Having both the AI ​​agent and the model processing everything makes it a much more competitive solution. Elon Musk has Cursor in his sights. Cursor’s good performance has led to growing interest in buying this company even before it becomes too big. Elon Musk knows this well and Grok, xAI’s model, is not so popular in the programming field. In April we learned that SpaceX had reached an agreement that gives you the option to buy Cursor for 60,000 million dollars. It would be a promising deal for both, because Composer 2.5 has already used Colossus’ infrastructure to train, and xAI could thus try to gain market share in the juicy enterprise sector. In Xataka | Elon Musk knows that TSMC is overwhelmed: Terafab is his idea to completely change the global chip industry

Sony has launched the most anti-2026 high-end mobile. It’s an idea as good as it is risky.

by

Sony continues launching mobile phones. And it has reached a point where the news is not that it has launched a mobile phone, but rather why it has done so and what it wants to tell the market when it does so. Although it may seem like a counter-current idea, launching mobile phones knowing that you are only going to sell them makes some senseand Sony is not alone there. The anti2026. For some reason, manufacturers have been convincing us for years that more than useful technologies should disappear. All in pursuit of a more minimalist design, larger batteries and an evolution close to that of the portless mobile. To this, Sony responds with a blunt “hold my cap.” The Sony Xperia 1 VIII. He Sony Xperia 1 VIII It is a return to the past, maintaining technologies that the vast majority of its rivals discarded years ago. It has a 3.5mm headphone jack It has a slot for microSD cards up to 2 TB Thick, very thick bezels The SIM slot is not removed with the tool, just press it The stereo speakers are front-facing, there is no one below There is no trace of what might look like an iPhone Very good. Sony’s proposal is clear: in the middle of 2026 you can enjoy technologies that one day we banished and that are useful despite the passing of the years. Furthermore, each and every one of these steps does not distance the Xperia 1 VIII from what is required of a modern flagship. The best Qualcomm processor A powerful camera system with ZEISS optics Up to a generous 16 GB of RAM The only unforgivable point is that of a Full HD+ panel. On a 1,499 euro mobile phone this resolution is not acceptable. Because. The short answer is that Sony is not launching the Xperia 1 VIII to compete with Samsung or Apple. He gave up that battle a long time ago. In the last Corporate Strategy MeetingSony president Kenichiro Yoshida made it clear that the smartphone division does not exist to sell volume, it exists to prove something. Sony Xperia is, above all, a technological showcase. In addition, the Xperia division is a fundamental pillar for the company’s R&D. The advances made in these smartphones are later applied to what really matters: the brand’s cameras. In Xataka | At half price the Sony WH-1000XM5, headphones with one of the best noise cancellations we have tested

The disastrous return of Amaia Montero with Van Gogh’s Ear is a good x-ray of the dangers of paying for nostalgia

by

Thirty thousand people have filled the BEC in Barakaldo for two nights to witness one of the most anticipated returns of Spanish pop. Amaia Montero returned to Van Gogh’s Ear almost twenty years later of his departure, on a tour called ‘So many things to tell’ and that promised to close a cycle (or open a new one). But these first concerts (especially the first one) have ended up being talked about for very different reasons than expected. The Ear, the return. When La Oreja de Van Gogh officially announced on October 15, 2025 that Amaia Montero was returning to the band, the tickets were sold out. in a matter of hours in numerous cities, and new dates were added due to demand. The tour consists of 16 stops that extend until November 2026 and include the Movistar Arena in Madrid (three nights), the Palau Sant Jordi in Barcelona and the Illunbe Donostia Arena. It is, right now, one of the most relevant pop phenomena of the year in Spain. How it started. The trigger It was an appearance that no one expected. in July 2024, by Amaia Montero at the Santiago Bernabéu during one of Karol G’s concerts to perform ‘Rosas’, the anthem from Van Gogh’s La Oreja. The reaction was enormous and shortly afterLa Oreja de Van Gogh announced the departure of Leire Martínez, the group’s vocalist for 17 years, from the band, alleging “different ways of living the group.” In October 2025, Amaia’s return was officially announced. 2026 is the year of the 30th anniversary of La Oreja and the 25th of ‘El viaje de Copperpot’, one of their most remembered albums. What happened in Barakaldo. On May 9, 2026, the tour started at the BEC and 18,000 people attended the first concert. Amaia appeared center stage on a raised platform, wearing a bright pink jumpsuit. He said: “I went down to hell itself, but with my scars, after fighting a lot, here I am.” However, despite good intentions, the chronicles they agreed in which Amaia was “out of tune and too tight to reach the high notes.” It especially went viral their performance of ‘Nothing Compares 2 U’the Prince classic popularized by Sinéad O’Connor and with which the band originally discovered her, but deeply out of tune. Amaia herself acknowledged on stage: “I do it terribly”. The amazing thing is that, as has also been saidAmaia “has had more than a year to prepare vocally” and despite this she showed “many technical deficiencies.” The group could have adapted the tonalities to their current voice, but they have not done so so that the songs are identical to how fans remember, and that is the problem. Second round. On May 10, in the second concert at the same venue, the setlist went from 25 songs to 22. ‘Nothing Compares 2 U’, ‘We’re all dancing to the same song’ and ‘The girl who cries at your parties’, the three songs that had generated the most problems the night before, disappeared. The presence of songs from Leire Martínez’s era was also reduced. Amaia was more confident and the elevated platform was reserved for ‘Mariposa’. The reactions of the fans. The acrimony with which Amaia’s return has been received has to do with the fact that the audience of La Oreja de Van Cogh is divided into two communities. The fans who came with Amaia, thirty-somethings who grew up with ‘Rosas’, ‘La playa’ or ’20 de Enero’, experienced the concert as the return of a diva and the disagreements were excusable. The fans who joined with Leire, on the other hand, listened to someone sing with obvious effort songs that his protégé performed for almost two decades with great technical solvency. A good part of the band’s fans (or ex-fans, at this point) think that there is a voice that has now been displaced, without a satisfactory public explanation and with an exit that It was tense: Amaia did not agree to share the 30th anniversary concerts with Leire, and the group chose to do without her. This has irremediably fractured the public, who in the case of the fans of the expelled vocalist, spread Amaia’s failures on social networks with particular viciousness. The comeback syndrome. Not all nostalgic reunions are a triumph. There are those that burst resoundingly because they try to revive an energy that responded to unrepeatable circumstances, but what they achieve is to make very clear just the opposite: that that energy no longer exists. The Sex Pistols on their 1996 tour is one of the most memorable cases for its self-awareness, but the drama dates back before: The Animals, Simon & Garfunkel or The Byrds in previous decades also failed when trying to win back their audience. The list of bands that recently crashed a reunion (not necessarily commercially, mind you) is endless: Jane’s Addiction, Guns’n Roses… The reason is always that the search for money is evident above the relationships between its members, sometimes very deteriorated. In this case, despite the fact that the economies of the band members They are very healthyfrom the videos we sense a certain suspicion and discomfort between the musicians and vocalist. Because that’s another: whether they succeed lyrically and commercially or not, what they won’t be able to escape is the evil tongues. As Leire fans know very well. In Xataka | The internal drama that Andy and Lucas lived through for years: story of a breakup that we are seeing live

More advanced chip factories are being built in China and Taiwan than anywhere else. It’s only good for them

by

According to SEMI, an international organization that looks after the interests of the electronics and integrated circuit industries, only six of the 64 new factories of semiconductors that are going to come into operation in Asia before 2029 will reside in Southeast Asia. The remaining 58 They will be located in China and Taiwan. These two countries have compelling reasons to strengthen its chip industry and develop its integrated circuit production capacity. It is essential for China to set up new plants equipped with cutting-edge photolithography equipment. And that is precisely what SMIC, Hua Hong Semiconductor and other Chinese chipmakers are doing. Currently this nation is limited by the difficulty of going beyond 7 nm without being able to use the extreme ultraviolet lithography (VVE) of ASML. Even so, Huali Microelectronics, the division of Hua Hong Semiconductor specialized in manufacturing chips for third parties, is preparing to start the production of 7nm integrated circuits at its Shanghai plant. Taiwan also needs to expand its semiconductor industry, although its motives are very different from China’s. The two largest Taiwanese integrated circuit manufacturers, TSMC and UMCthey need to develop more cutting-edge plants in order to satisfy the growing needs of their customers. TSMC’s 2 and 3 nm nodes in particular cannot cope, so it is essential for this company to expand its production capacity in the midst of the boom in data centers for data applications. artificial intelligence (AI). SEMI is concerned about the vulnerabilities of the chip industry Ajit Manocha, the executive director of SEMI, assures that “we want to see more centers emerge in related countries. We want more plants to be established to reduce the risk derived from vulnerabilities.” What worries the spokesperson of this organization is that the geopolitical tensions maintained by the US, China and Taiwan end up threatening the integrated circuit factories that reside in these last two countries. TSMC’s in Taiwan are especially sensitive to a possible conflict with China due to the undoubted strategic importance that they have not only for Taiwan, but also for the US and its allies. Malaysia, Singapore, Vietnam and Thailand are candidates to host new cutting-edge chip plants Malaysia, Singapore, Vietnam and Thailand are strong candidates to host new cutting-edge chip manufacturing plants. In fact, Several centers already reside in Malaysia Intel’s advanced packaging and verification software. However, Manocha You are also concerned about other types of vulnerabilities. The most critical of all is the shortage of critical minerals, as well as bromine and helium, two fundamental gases in chip manufacturing processes. What is happening with helium in particular is very worrying. This gas is a byproduct of natural gas processing, and its price skyrocketed in March shortly after the war that the US, Israel and Iran have been fighting since then began because Qatar was forced to stop production of liquefied natural gas. In the current unstable scenario, SEMI argues that Southeast Asian countries should aim to build more semiconductor manufacturing plants over the next decade to help the sector diversify and reduce supply risks. Image | TSMC More information | Reuters In Xataka | The US’s problem in the AI ​​and humanoid race is not China: it is all of Asia and it is greatly disadvantaged

Steve Jobs’ widow is squandering the fortune she inherited. You have good reasons to do so.

by

Laurene Powell Jobs was already a brilliant business woman with a promising future before she met, almost by chance, the person who would be her life partner for the next 22 years. As chance would have it, one day he arrived late to a conference in 1989 and sat next to the main speaker: Steve Jobs. What happened from that moment on is part of the history of technology. After the death of jobsLaurene inherited much of the Apple founder’s fortune, which she only had to share with Steve’s first daughter, Lisa Brennan-Jobs. Lisa, in addition to giving the name to the failed Apple projectwas the result of Jobs’ relationship at age 23. Most of Laurene’s inheritance was Jobs’ shareholding in Apple and Disney. In 2011, these shares were valued at around 10 billion dollarsbut Jobs’ widow was very clear about what she was going to do with that money: “I am not interested in building on the legacy of an inheritance and my children know it. Steve was not interested in that. If I live long enough, that inheritance will end with me.” The latest movements of Laurene Powell Jobs indicate that Jobs’ widow will keep her word and spend the entire fortune she inherited as Steve would have liked: dedicating herself to philanthropy until her death. Goal: donate $3.5 billion over the next 10 years Just like MacKenzie Scottex-wife of Jeff Bezos, Laurene Powell Jobs did not want to join her multimillion-dollar charity project to the The Giving Pledge Philanthropic Clubled by Bill Gates and Warren Buffett. He has preferred to go it alone and spend his entire fortune on his own philanthropic projects that improve people’s lives and reduce the impact of climate change. Graduated in political science and with a master’s degree in business from Stanford, the millionaire widow She is not a novice managing funds.. In addition to creating his own healthy eating company, Powell Jobs has been able to make impact investments which have allowed it to support social and environmental projects without its fortune being significantly reduced. The value of Disney and Apple shares has increased tenfold in the last decade, however, Laurene’s current fortune is estimated at around $14 billion. That gives an idea of ​​the volume of donations he has made in recent years. Since the death of Steve Jobs, Laurene has created two charitable foundations on which she concentrates all her philanthropic efforts. The first is Emerson Collective which focuses on educational projects that seek to offer equal educational opportunities for groups at risk of exclusion. The second pillar of your charitable project is Waverley Street Foundationan international initiative aimed at protecting the groups most vulnerable to climate change, supporting education, health and preservation projects of natural spaces so that people can survive in the communities where they were born. Jobs’ widow’s project is invest 3.5 billion dollars in the next 10 years in this latest project. “I inherited my wealth from my husband, who didn’t mind accumulating it. I do this in honor of his work and have dedicated my life to doing everything I can to distribute it effectively, helping people and communities sustainably.” This investment objective is far from 10 billion dollars that Jeff Bezos proposed to investthe $45 billion from Mark Zuckerbergthe most 160 billion from Warren Buffet or the 90% of Bill Gates’ fortune. Laurene’s philanthropic spirit and discretion does not prevent the widow from enjoying her fortune visiting Mallorca on the Venusa family yacht designed in 2009 by Jobs himself, valued at 120 million dollars. Jobs’ fortune does not concern the couple’s three children either, since all of them already have established careers outside the media spotlight. Reed Paul relegated his father’s last name to the background to pursue a degree in oncology at Stanford University. Erin Siena is an architect and designer. Eve Jobs has a degree in Science, Technology and Society from Stanford, although she currently makes her living as model on the main catwalks. In Xataka | “It’s not what you say, it’s how you say it”: Steve Jobs’ technique that used emotional intelligence when no one was talking about it In Xataka | It’s not Steve Jobs, it’s Mustafa Suleyman: Microsoft’s AI CEO who joins the trend of dressing “Jobs style” Image | Flickr (TechCrunch)

Two men thought it was a good idea to lend their houses to a North Korean laptop farm. It went wrong

by

Teleworking has accustomed us to a very comfortable idea: if someone delivers work, attends meetings and responds to messages, perhaps it doesn’t matter too much where they do it from. The problem appears when that distance becomes an advantage to hide identities, move money and enter companies that believe they are hiring a legitimate professional. North Korea has been exploiting precisely that rift. And the case of two men convicted of hosting laptops in their homes shows the extent to which the plot could rely on domestic infrastructure. Two men condemned. Matthew Isaac Knoot, of Nashville, Tennessee, and Erick Ntekereze Prince, of New York, have been sentenced in the US to 18 months each in prison for their role in fraudulent schemes involving remote IT workers linked to North Korea. according to the Department of Justice. The house as a piece of the plot. The mechanism was more domestic than one might imagine. Companies shipped corporate laptops to American addresses because they believed the contracted workers were there. Once received, the computers were housed in those homes and configured with remote desktop applications installed without authorization. This allowed the fake workers to operate from abroad while, to the companies, the connection appeared to come from an address within the United States. What did each one do?. Prince, according to official information, facilitated at least three North Korean IT workers to obtain remote employment in US companies between June 2020 and August 2024, and used his company Taggcar Inc. to fraudulently supply “certified” workers, despite knowing that they were outside the US and using false or stolen identities. Knoot, for his part, operated a laptop farm from his Nashville residences between July 2022 and August 2023. Money, companies and damages. The Justice Department maintains that the two schemes together generated more than $1.2 million for North Korea and affected nearly 70 U.S. companies. In the Prince case, the companies paid more than $943,069 in salaries to IT workers linked to the file. In Knoot’s case, the payments exceeded $250,000. More than labor fraud. The US justice system presents the sentences as part of a specific line of action against facilitators located in the US. The note itself highlights that these are the seventh and eighth convictions of “laptop farmers” obtained in the last five months within their efforts to interrupt North Korea’s illicit generation of income. It is an important nuance: the focus is not only on those who connect from abroad, but also on the local network that makes the operation viable. Expansion into Europe. As we have seen in the pastthese cases are also present outside the United States. The Record discovered in April 2025 an investigation by Google Threat Intelligence Group according to which North Korean operatives had increased their activity in Europe following US police actions against laptop farms and financial networks. At the center were job searches linked to the United Kingdom, Germany and Portugal, in addition to the use of local facilitators to support the alibi of a work presence in the corresponding country. AI and fake identities. One of the most current layers of this story is not only in the laptops, but in the ease of building increasingly credible profiles. BISI points out that North Korean operations combine stolen identities, manipulated professional profiles and AI tools capable of writing localized CVs and cover letters. In the Old Continent, platforms such as Upwork and Freelancer are usually used, in addition to Telegram. The consequence is obvious: detecting the fake candidate can become much more difficult before the company even ships the equipment. What started with laptops housed in private homes ends up having something much bigger than a criminal conviction. The companies were not attacked from outside in the classic sense, but ended up opening the door to workers they believed to be legitimate. So everything seems to indicate that in these times it is no longer enough to protect servers, credentials or repositories, but rather to review the processes that we consider normal, such as the hiring of personnel. Images | Xataka with Grok In Xataka | The ‘vibe coding’ promised to democratize software. Your first gift is 5,000 apps with open sensitive data

The EU said its AI Law was prodigious. It must not be so good when it is going to delay its application until 2028

by

Brussels has agreed delay the toughest restrictions of your AI Law until December 2027, which in practice moves its real impact to 2028. Initially the calendar was much more ambitious and intended to prohibit and punish AI systems classified as “unacceptable risk”but all that now remains a dead letter. This makes it clear that community institutions are not prepared to supervise what they intended to regulate. Tell AESIA. A victory with risks. German Chancellor Friedrich Merz has made Germany dictate the pace of Brussels. Mez has pushed until the last minute to get industrial AI applications virtually off the radar of the law. This will allow German business giants such as Siemens or Bosch not to have to comply with the regulations. Is according to Politico a political victory for Berlin—although other companies like ASML asked for the same— which protects its heavyweights, but poses a problem: if we let AI control factories and critical infrastructure, we risk its failures having an enormous impact, which is exactly what the AI ​​Law proposed. What was considered high risk six months ago is no longer so. Brussels comes to its senses. MEPs have understood something that they had been refusing to admit for a long time: being the “sheriff” of the internet is of no use if you don’t have your own AI industry. While Europe regulated, the US and China they grew without brake. The agreement is the clear admission that regulating with a heavy hand a market that you do not dominate does not make sense. The EU now gives its companies some oxygen instead of forcing regulations that the rest of the world is simply ignoring. The “Brussels effect” has a ceiling, and this delay marks it. The AI ​​Law does not give up completely. The EU, however, has included an express prohibition against AI systems capable of generating deepfakes of recognizable people. It is a direct response to the controversy generated by tools like those present in Xand a way to keep the protective spirit of the law somewhat alive. The obligation to identify AI-generated content also remains, but the grace period is stricter now, at three months compared to the previous six months. Even so, there is a clear surrender in what mattered most. Careful. If the EU has decided to delay its flagship AI law in the face of industrial pressure, what will prevent the DMA or the DSA from ending up suffering the same fate? The two regulations have been involved in complex industrial battles for some time: Apple and Meta they continue to resist to meet the requirements of interoperability and transparency, and the Commission has had to qualify its own requirements. The precedent of the is dangerous because it shows that political pressure works. Regulate so much for what. The EU has been wanting to lead technological regulation without leading (or even competing in many areas) in technological innovation. The GDPR served as a global standard because Europe was a large enough market to impose conditions for entry. The difference is that AI depends on something else very different, and here the feeling is that the only thing Europe is doing is putting doors on the field. Let your own AI Law end modifying first and later delaying is nothing more than a tacit recognition that the EU’s regulatory strategy and ambition has been a shot in the foot. One that has worsened the conditions to be able to compete with those of other countries. Image | World Economic Forum In Xataka | “What is allowed in China can never be allowed in Europe”: Spanish MEPs from the AI ​​Special Commission speak

50 years later, the Soviet fire of the “Gates to Hell” is going out. And it’s not good news

by

In 1971, in the heart of the Karakum Desert, a group of Soviet engineers observed how the ground was sinking under his feet after a failed drilling. What came next was not an immediate evacuation or closure of the area, but rather an improvised decision that, according to who witnessed itseemed like a quick solution to a specific problem. That choice, taken almost as another technical procedure, would end up having consequences that no one at that time was able to anticipate. The eternal fire goes out. During more than half a centurythe Darvaza crater has burned relentlessly in the middle of the desert, becoming an almost permanent image of inexhaustible fire that seemed to defy any natural logic. However, the most recent data show a clear change: the intensity of the flames has fallen drastically in recent years, losing more than 7% of its strength. What for decades was a constant spectacle begins to weaken, altering the perception of a phenomenon that many considered eternal. The origin between legend and Soviet heritage. The birth of the crater is still shrouded in all kinds of stories and uncertainty, although the most widespread and feasible version points to the accident. during Soviet drilling in search of gas in the sixties or seventies. According to this theory, the ground collapsed when it reached a pocket of natural gas and the engineers they decided to set fire to the site to prevent the release of toxic gases, convinced that it would be extinguished in a short time. Thus, what was going to last weeks lasted for decadesfed by an underground network of gas that never stopped flowing, giving rise to one of the best-known anomalies of the energy legacy of the former Soviet Union. From remote curiosity to global icon. Over time, the crater went from being a geological oddity to becoming a almost mythical destiny for travelers and explorers, despite the difficulties in accessing Turkmenistan. Its image, a gigantic burning cavity in the middle of nowhere, has fueled so much adventure tourism like internal propagandato the point of being used by country leaders as a symbol of power or control. The experience of approaching the edge and feeling the direct heat of the fire has reinforced its reputation as a unique place in the world. The attempt at control and doubts about its decline. For its part, the Turkmen government has years trying to control emissions from the crater, and attributes part of the recent weakening to new drilling nearby plants intended to extract gas. However, the independent analyzes They suggest that the loss of intensity could have begun before these interventions, which opens the door to natural causes that are not yet fully understood. This nuance introduces a key and dangerous uncertainty: it is not clear whether the end of the phenomenon responds to human action or to a change in the geological system itself. The unexpected twist: less fire does not mean less problem. Yes, because although At first glance, the reduction of the flames could seem like good news from an environmental point of view, the reality it is more complex. Fire acts as a mechanism that transforms methane (much more powerful as a greenhouse gas) into carbon dioxide, reducing its impact in the short term. If the flames subside, more methane could be released directly into the atmospherewhich would make progressive shutdown a potentially bigger problem. A fragile balance that is still active. Despite its weakening, the crater remains activewith visible flames and constant emissions that remind us that the phenomenon has not disappeared. The huge amount of gas accumulated underground suggests that the fire will not be completely extinguished in the short term, maintaining that strange balance between natural spectacle, industrial legacy and environmental problem. Thus, half a century later, the symbol of eternal fire begins to change, although its disappearance does not necessarily imply a more favorable end for the rest of the planet. Image | Stefan Krasowski, Tormod Sandtorv In Xataka | China’s first pipeline network is 4,000 years old and something revolutionary: it was built without the need for kings or nobles In Xataka | About to close, this remote mine in the Polar Circle has found a 2 billion-year-old yellow diamond that weighs 158 carats

Mistral has a new AI model. The good news is that it is absolutely European; the bad one, which is absolutely mediocre

by

The French startup Mistral has just launched Mistral Medium 3.5an open-weight AI model that is the great European exponent in an industry absolutely dominated by China—which competes directly with this type of projects—and by the US. And if this is the best they can do, it seems Europe has a problem. Mediocre. This is a “dense” model with 128 billion parameters and a context window of 256,000 tokens. While models with Mixture-of-Experts (MoE) architecture only activate a subset of the total parameters to achieve enviable efficiency and capacity, Mistral activates them all. That makes it much less efficient, but theoretically it should make its performance promising. And that’s the problem. Which it is not. Benchmarks. Pedro Domingos, professor of deep learning at the University of Washington, he expressed it very well: “Mainstream AI companies brag about how their model is much better in benchmarks. Soo Mistral brags about how their model is much worse.” It is true that the models with which it is compared are larger in total number of parameters, but as we will see later, even taking that into account, they are cheaper and theoretically more efficient thanks to the use of that MoE architecture in many of them. The model, however, unifies the previous catalog and follows the market trend of being able to establish the desired level of reasoning (reasoning_effort) as a parameter. Bad results. And he is somewhat right: Mistral does not seem to have problems showing the results of various benchmarks in which it performs poorly, but it also performs poorly with models that are by no means the most recent or powerful on the market. Thus, it is compared with Claude Sonnet 4.5/4.6, with Kimi K2.5, with GLM-5.1 or with Qwen 3.5 397B. In almost all cases (except GLM 5.1) there are already more recent and powerful versions of all of them. Not so far from local models. In fact Medium 3.5 scored 77.6% in SWE-Bench Verified, a programming test in which Qwen3.6-27b It reaches 72.4% with a fundamental difference: you can run it “for free” (with the appropriate hardware, and you paying the electricity bill) with a relatively affordable machine. More expensive (and somewhat more restrictive). If we use it via API, Mistral Medium 3.5 costs $1.50 per million input tokens and $7.5 per million output tokens. GLM-5.1 costs 1.4/4.4 respectively, and Kimi K2.5 costs 0.5/2.8 respectively. Its recent successor, Kimi K2.6, costs 0.95/4, and it is significantly better than Mistral being cheaper. There is a curious fact: Mistral uses a “modified MIT license” instead of the traditional Apache 2.0, and indicates that this model can be used commercially or non-commercially except for “high-income” companies. Chasing Anthropic. In addition to the model itself, the company has presented the so-called remote scheduling agents using Mistral Vibe CLI to, for example, send pull requests to GitHub in an automated way. It also has the so-called “Work Mode” in LeChat, allowing multi-step tasks to be managed autonomously. These are tools clearly intended to strengthen Mistral’s role as a base for scheduling agents, which is the path that has worked fantastically for Anthropic. Your advantage: being European. The only great strength of this model is that it has been developed by a European startup, and that gives it clear visibility at a time when many EU countries they talk about digital sovereignty. It is the only Western model that seems to want to compete with China in the field of open weight models, which is good news, but the truth is that in terms of performance it does not seem that the Mistral Medium 3.5 is going to perform competitively. The geopolitical security network. That, together with the fact that it costs more than its competitors, makes the decision to use it difficult unless for those who prioritize clearly that European origin. That is Mistral’s ace in the hole, and they are taking advantage of perfectly. The company has recently obtained financing to create data centers in Europe, and is nourished and fed by this new obsession with minimize dependency of North American Big Tech. In Xataka | The CEO of Mistral sends a message to Europe: the end of being the technological vassal of the United States

In the middle of the RAM crisis, your cheapest computer was a bargain too good to last

by

If there is a product in Apple’s portfolio that was a real candy, it was the Mac Mini. This has been a reality for years, but in these times that are even more so: the Mac Mini M4 It came to the market with the power of the M4 chip, 16 GB of base RAM, a 256 GB SSD (the most stingy, Apple style) and a RRP of 719 euros, which in practice was much less. I bought it myself for less than 600 euros. Well, that bargain has come to an end: in the context of current shortages, the 256GB Mac Mini is no longer an option. We had already seen it with its models with more RAMbut this decision is dramatic for the general public. Goodbye to the 256 GB Mac Mini. Apple has made a decision that directly affects the pockets of those who want to buy the Mac mini in its most basic version. Since yesterday, May 1, 2026, Cupertino has removed that entry model from its catalog, as Joe Rossignol advances for MacRumors. It is not that it appears out of stock, it is that it has directly disappeared, as can be seen on the Spanish website. Of course, there is still stock and offers of the old base model in stores like amazon, at Media Markt either in El Corte Inglés. The entry price of the Apple desktop computer starts at 969 euros and corresponds to the version with M4 chip, 16 GB of RAM and 512GB SSD. In the United States the jump has been from 599 dollars to 799 dollars, in Spain it has gone from 719 euros to 969 euros. The versions with the M4 Pro processor remain as they are. This decision is framed within a structural RAM supply crisis and whose main culprit is the voracity of the AI ​​infrastructure. Prices and delivery for the Mac Mini. Apple Why is it important. Raising the entry price of one of its star products by almost 35% (in the United States it is 33) more is an aggressive move that has implications for both the individual consumer and the technology market in general. It is true that technically speaking Apple has not raised its prices, it has simply eliminated the lower step, leaving orphans those people who considered that base version sufficient, which are quite a few: it is my main computer for mixed tasks, basic editing, office automation and the Internet and the performance is more than good. In short: for many users, students or professionals, with tight budgets, this increase of more than 200 euros is a real chore. The problem is not just the price: the impact is worsened by delivery times. I have tried different Apple Stores and shipping is delayed until the end of May or beginning of June. Context. Tim Cook gave an explanation during the conference results for April 30, 2026 recognizing that the supply of Mac mini and Mac Studio is severely restricted and that normalization could take months. The reason given by the still CEO of Apple is that both devices have become popular platforms for artificial intelligence and agentic tools, which has triggered demand above forecasts. And he anticipated something: Apple will face significantly higher memory costs in the current quarter, according to MacRumors. This places the Mac Mini in a paradox: that the configuration of this compact desktop computer makes it ideal for working with AI locally and that precisely this reality is what has exhausted the stock, forcing Apple to cut its catalog. The AMR crisis continues to claim victims. In March of this year Apple already removed the 512 GB RAM option from the Mac Studio and in April several models of the Mac mini and Mac Studio they directly stopped being able to order in the Apple Store in the United States, with delays of up to five months for versions with more RAM. The memory chip supply crisis is not something exclusive to Apple, but a trend that crosses the entire sector and caused by the demand of the hyperscalers. Apple needs to ensure that every machine sold is capable of fluidly running its new digital agents and AI tools, making lower memory and storage configurations no longer viable or cost-effective under the company’s current standards. The particular thing about Apple’s decision is the timing: just when it launches its best chips for local AI processing, the global RAM market is strained to unsuspected limits precisely because of that fever. The result is paid by the final consumer. In Xataka | Not even Apple is free from the new reality of the technology industry: RAM goes first for hyperscalers In Xataka | The RAM crisis was supposed to make computers and smartphones very expensive. Apple has another opinion Cover | Apple and Alberto García

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.