Anthropics

list of new features of the new version of Anthropic’s Artificial Intelligence model

by

Let’s tell you What’s new about Claude Fable 5, Anthropic’s new public model for your chat artificial intelligenceThis is the first model of the new Mythos classwhich already when it launched its preview version two months ago did so saying that it was so powerful that they would limit its power. Anthropic has launched two new models, the Claude Mythos 5 which will be available only to “a small group of cyber defenders and infrastructure providers”, and a Claude Fable 5 more adapted for mainstream users. It is a model with all the capabilities of Mythos Preview, but with security measures so that it is not used for bad things. And precisely because it is the model that is reaching users, it will be precisely Claude Fable 5 that we are going to focus on. we will give you a list with everything that changes so you know what it’s like to use compared to the previous version. Claude Fable 5 news Mythos-level capabilities for everyone: Fable 5, along with Mythos 5, is the most powerful AI model in history. Their capabilities are enormous, with performance that makes them lead the test benches. It is the first that allows you to use the capabilities of Mythos 5, and the jump from Claude Opus 4.8 is truly surprising, leaving GPT-5.5 and Gemini 3.1 Pro far behind. Longer freelance work: Fable 5 can work autonomously for longer than their predecessors. This makes it possible to tackle longer and more complex tasks that could not be sustained before. A big leap in programming: Programming is another aspect where this model has improved the most. Anthropic highlights the case of Stripewhich during initial testing performed a complete migration of a 50 million-line Ruby code base in one day, a job that would have taken a human team more than two months. More efficiency: Fable 5 is also more efficient in token consumption than previous Claude models, good news for teams who want to get the most out of it. Improvements in knowledge work: Fable 5 shows solid performance on complex analytical tasks. Anthropic claims to outperform the competition in senior-level reasoning, with notable improvements in reasoning about documents, interpreting graphs and tables, and problem solving. Great improvement in vision tasks: This new model can extract concrete figures from detailed scientific figures and reconstruct the source code of a website based solely on screenshots. For example, he was even able to beat Pokémon FireRed based on vision alone, while previous models had trouble playing even with assistive tools. Better memory and long context: The model is able to not get lost over millions of tokens, maintaining focus on the context, and improving memory. Something perfect for longer and more complex tasks. Maintains its barriers against misuse: Anthropic also says its new model keeps the level of misaligned behavior. This means that deceiving him or trying to get him to cooperate in misuse is still just as difficult. Safeguards with forwarding to Opus 4.8: Fable 5 is based on Mythos 5, which Anthropic said was so powerful it was scary. That is why it comes quite well-equipped and with many security measures. For example, if it detects that we are asking something “dangerous”, it avoids the question and even forces the use of an inferior model, Claude Opus 4.8. New data retention policy: For Mythos-class models, including Fable 5, Anthropic requires 30-day data retention for security monitoring purposes, both on its own and third-party surfaces. Of course, they say that they will not use that data to train new models or for anything that is not related to security. After 30 days, it will delete all data from our conversations in almost all cases. Price and availability: Fable 5 costs $10 per million tokens in and $50 per million tokens out, less than half of what Claude Mythos Preview cost. Developers can use it in the Claude API with the identifier claude-fable-5. Plus, it’s included at no extra cost until June 22 for Pro, Max, Team, and Enterprise subscribers. Of course, starting June 23, it will require usage credits until the capabilities of the company’s servers allow it to be incorporated as standard in payment plans in the future. In Xataka Basics | How to prevent AI from always being right by default and thus make Claude, Gemini and ChatGPT have fewer hallucinations

Anthropic’s AI already writes 80% of its own code because it was inevitable that AIs would improve themselves

by

“As of May 2026, more than 80% of the code we integrate into the Anthropic codebase was created by Claude.” Those who reveal this information are two Anthropic researchers who have published one of the most revealing texts about the present and future of the company’s AI models. One that tells us about a fascinating and disturbing concept in equal parts called recursive self-improvement. Code multiplier. The impact of these agentic programming tools on the work of Anthropic engineers is being spectacular. According to internal Anthropic data from May 2026, this autonomous code generation has caused an Anthropic engineer to produce eight times more lines of code per quarter today than during the 2021-2025 period. Anthropic’s human programmers they no longer program– Direct and review AI-generated code. A frenetic evolution. The changes we have experienced have been fascinating, they explain in Anthropic. Between 2021 and 2023, engineers wrote all code by hand on their computers. In 2024 they started using chatbots to generate small snippets of code that they then copied and pasted. In 2025, agents capable of work autonomously on entire files. Longer time in a row. According to the METR benchmark which measures the ability of AI to complete complex tasks, in 2022 GPT-3.5 could barely last about 35 seconds operating autonomously without making serious errors. By mid-2026 Claude Opus 4.6 is already capable of working 16 hours in a row on complex tasks. At Anthropic they point out that the length of tasks that an AI model could undertake doubled every seven months, but now it doubles every four. If this trend continues, “tasks that take a person days could be automated with AI. By 2027, AI systems could be able to work on tasks that take a person weeks.” Superhuman performance. Industry benchmarks are being “saturated” by new AI models, which already reach almost 100% of the possible score in many of them. For example SWE-bench, which measured the models’ ability to program, was almost is surpassed for the most recent models. In 2025 Claude pus managed to optimize the code they gave him by making it ran 3x faster. In April 2026 Claude Mythos Preview already achieved a 52x speedup of that code. AI that improved itself. This concept of recursive self-improvement presents a scenario in which an AI model generates data, corrects its own failures, and trains itself continuously. This opens the door to exponential growth in its capabilities, but at the same time reopens a debate on the risks that this type of evolution generates. Source: Anthropic infinite loop. Traditionally, human engineers analyzed the responses of a model, cleaned the data, and adjusted parameters to create the next version of that model. With recursive self-improvement AI takes on that role and evaluates its own performance, generating more complex problems to test itself and generating synthetic data for your next generation. Danger. This autonomy implies a potential risk: that humans lose control of where the AI ​​goes. That we do not know or can assure if it is aligned with our ethics and ideals. The biaseshowever small, can be amplified with this type of iterative process, but the model itself may have mutated its original ethical reasoning mechanisms and security protocols to become something totally unpredictable. The Terminator scenario. Isolation and arbitration. To avoid these risks, at Anthropic they implement this evolution in isolated environments to then verify that everything works as it should. In addition, the company uses independent evaluation models that act as independent arbiters that audit these models. that evolve by themselves. They do this by checking each change in the code to prevent its impact from being harmful to the system or to those who use it. The new bottleneck is the human being. The Amdahl’s law is a formula that is used to find the maximum performance improvement of an information system when only a part of that system is improved. At Anthropic they point out how as AI continues to write more and more code, the real bottleneck is the human being who has to review that code. In Xataka | Anthropic is one step away from being worth as much as Samsung. And what the market is buying is not Claude

list of new features of the new version of Anthropic’s Artificial Intelligence model

by

Let’s tell you What’s new in Claude 4.8 Opus, the new version of Anthropic’s most advanced and powerful artificial intelligence model. This version has surprised us by arriving just 41 days after Claude Opus 4.7, and it seems that the improvements are minimal, but there is a really important change in its honesty when it comes to telling you if it doesn’t know something. In any case, here you have a complete list with all the new features that come with this new version of Claude 4.8 Opus. We are going to explain each of them briefly so that they are easy to understand. Another thing you should know is that Opus is the most advanced line of Claude models, the one indicated for more complex tasks for programming and the one that uses up your limits the fastest when you use it. There is also the most efficient Sonnet model for day-to-day tasks, which continues in version 4.6 since February 2026, and a Haiku for quick and simple questions that continues in version 4.5 since October 2025. News from Claude 4.8 Opus A more honest AI: The prominence of this new version goes to honesty. He’s significantly more honest about his own work, telling you when he’s unsure about something. It’s also about four times less likely to let bugs in code slip by without flagging them, compared to its predecessor. Performance improvements: The Agentic code score for creating code with agents increases from 64.3% to 69.2%, and multidisciplinary reasoning with tools increases from 54.7% to 57.9%. On other test benchesin the SWE-bench Verified it goes from 87.6% (Opus 4.7) to 88.6%, and in Terminal-Bench 2.1 it rises from 66.1% to 74.6%. GPT-5.5 still falls short in terminal/CLI workflows, although there have been big improvements in Claude, and both models are practically on par in web browsing and graduate-level science topics. Alignment improvements: Alignment assessments show new highs in prosocial traits such as supporting user autonomy and acting in their best interest. Rates of misaligned behavior such as cheating are lower than in Opus 4.7. Fewer hallucinations: As usual, the number of hallucinations is also reduced. Honesty when telling yourself when you don’t know something also helps reduce them. Quick mode: According to AnthropicOpus 4.8’s fast mode is now about 2.5 times faster. The company claims that the improved Quick Mode also costs three times less than before. Effort control– Users can choose between “extra” or “max” levels so that the model spends more tokens and obtains better results. Dynamic Workflows (preview for research): With this new feature, Claude can schedule work and run hundreds of subagents in parallel in a single session, being able to complete codebase-scale migrations of hundreds of thousands of lines. Available for Claude Code on Enterprise, Team and Max plans. No change in base price: The base price of API tokens is unchanged from Opus 4.7. It is 5 dollars per million input tokens, and 25 dollars per million output, with up to 90% savings with prompt caching and 50% with batch processing. In Xataka Basics | How to prevent AI from always being right by default and thus make Claude, Gemini and ChatGPT have fewer hallucinations

Anthropic’s security manager leaves the company to write poetry

by

In a movement more typical of “nihilistic penguin“that the head of security for one of the main protagonists in the development of AI, Mrinank Sharma, head of artificial intelligence security at Anthropic, has announced his resignation with a public letter in your X profile and he will dedicate his life to writing poetry. In his statement, Sharma not only explained why he is leaving the company that develops the models of Claudebut instead described the current state of AI development, with language that mixes alarm with personal reflection. “The world is in danger,” said the former director of Anthropic. The context: who he is and what he did at Anthropic. Mrinank Sharma headed the Safeguards Research Team from Anthropic, a research group focused on studying the risks associated with AI systems. Within Anthropic, Sharma’s work included developing defenses against risks such as AI-assisted bioterrorism and studying phenomena such as sycophancy (the tendency of AI models to user adulation), as well as investigate how AI can influence human perception and change cultural behaviors. He leaves, but leaves a message. The almost cryptic letter that Sharma published in X It quickly went viral due to the messages it contained. In it, he expressed his concerns in a tone that transcends the technical. One of the quotes that has attracted the most attention: “The world is in danger. And not only because of AI, or biological weapons, but because of a series of interconnected crises that are developing at this very moment.” Beyond the almost apocalyptic literalism, Sharma warned that humanity was approaching a critical point in which the development of AI was facing ethical dilemmas for those who develop it “our wisdom must grow at the same rate as our ability to affect the world, otherwise we will face the consequences.” Work to stay out of work. Sharma is not the only one who faces this ethical dilemma. According to sources of The Telegraphother Anthropic employees have expressed concern about the huge evolutionary leap in the latest AI models. “I feel like I come to work every day to stay out of work“one of the employees acknowledged to the British media. In a way this is true, since these employees are working on the development of a technology that, in all likelihood, change nature of his work, and that of millions of peoplea few years away. Is that good or bad? A first reading of the letter leaves the feeling that these workers are developing the weapon that will destroy humanity. However, a reading between the lines leaves Anthropic in a pioneering situation compared to its rivals from OpenAI, Microsoft or xAI: they are achieving advance at a pace which overwhelms even its developers. A sensation that does not seem to occur in the templates of other companies. Could it be that their models are not at that point of evolution? “Throughout my time here, I have seen repeatedly how difficult it is to allow our values ​​to guide our actions. We constantly face pressure to let go of what matters most,” Sharma wrote. The poetic turn. In addition to reflecting on the global risks he perceives, Sharma announced that his next professional step will be very different from the one he had until now. In his letter he mentioned his intention to devote time to what he called “the practice of courageous speech” through poetry. This change of lA for poetry has been interpreted as a sign of dissatisfaction with the pace and focus prevailing in the AI ​​technology industry. Like Sharma, in recent weeks other key figures in Anthropic’s AI development have announced their resignation. Harsh Mehta and Behnam Neyshabur They also announced a few days ago that they were leaving the company. However, in these cases, the exit announcement was made and, immediately afterwards, a new AI project was announced. That is to say, far from the ethical postulates that Sharma proposed, his intention was more along the lines of digging into his own gold mine and not that of others. In Xataka | Daniela Amodei, co-founder of Anthropic: “studying humanities will be more important than ever” Image | mrinank sharmaAnthropic

“They are brilliant researchers under the control of an authoritarian government.” Anthropic’s CEO has spoken about Depseek

by

In the midst of the stir caused by the latest models of the Deepseek, the CEO of Anthropic, Dario Amodei, has published An analysis on its personal website in which it questions the narrative of the “Chinese miracle” in artificial intelligence. Why is it important. The debate on Chinese capacity to develop advanced AI has monopolized the agenda in recent days after Deepseek’s releaseswhich have come to provoke A 17% drop in Nvidia shares. The facts. Deepseek claims to have developed its model V3 for just under 6 million dollarswhile Amodei explained that Claude 3.5 Sonnetthe last and most advanced Anthropic model, required “some tens of millions” in training. Far from the “thousands of millions” that were speculated. “Deepseek has produced a model close to the performance of US models 7-10 months ago, for a rather lower cost, but not in the proportions that have been suggested,” said the CEO. Deepseek operates with about 50,000 generation chips Hoppera capacity that Amodei considers similar to that of the main American technological ones. According to his analysis, Deepseek’s advances reflect the natural reduction of costs in the sector, estimated at annual 75%. The context. Deepseek has presented two models: V3, which uses traditional training. And R1, which incorporates reinforcement learning. For Amodei, real innovation is in V3, not in R1, which according to him, follows roads already explored by other technological ones. Turning point. The development of an AI superior to human intelligence will require millions of chips and tens of billions of dollars in the coming years. “Between 2026 and 2027 we will see which will be smarter than almost all humans in almost all tasks,” he said. In this scenario, he has defended Export controls as a strategic tool. Amodei has also recognized the talent of Deepseek engineers … although he has warned about the implications that a company operates under the control of the Chinese government. For him, The growing efficiency In the development of AI justifies reinforce, and not relax, commercial restrictions. In fact, he has had some words of praise for Depseek’s team, but not for his nation: “They are brilliant and curious researchers who only want to create useful technology, but are subject to an authoritarian government that has committed human rights violations and He has behaved aggressively on the world scene. “ In Xataka | “Google gives you links, perplexity gives you answers”: we talk to the CEO of the startup that wants to kill the father Outstanding image | Techcrunch

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.