Anthropic is considering launching a new flagship AI model following OpenAI\u2019s rapid expansion in the corporate market. GPT-6 Astra, unveiled in early September, is rapidly gaining clients and has already prompted potential investors in Anthropic to take a closer look at the company\u2019s position ahead of a possible IPO. The paradox is that all this is happening just as Anthropic\u2019s CEO, Dario Amodei, is calling on the industry to slow down the development of the most powerful systems due to the risks posed by AI.<\/p>\n
Key points at a glance:<\/strong><\/p>\n Back in the summer, Anthropic\u2019s position looked exceptionally strong. Claude had become one of the key tools for programming, corporate analytics and complex intellectual work. The company was rapidly increasing its revenue, and its models were being used more and more extensively by large business clients.<\/p>\n 3 September OpenAI<\/a> has released GPT-6 Astra <\/strong>and the balance began to shift.<\/p>\n OpenAI positions Astra not simply as a new chatbot model, but as a system designed for complex professional tasks. It is geared towards programming, browser-based tasks, computer-related work, cyber security, science and multi-step autonomous processes.<\/p>\n This is precisely what is putting direct pressure on Anthropic. Over the years, Claude has grown stronger in precisely those areas where businesses see the greatest practical value: writing code, working with large amounts of data and performing complex tasks with minimal human intervention. Now OpenAI is targeting that very same segment.<\/p>\n Following its launch, GPT-6 Astra quickly gained popularity amongst corporate users.<\/p>\n The situation currently looks like this:<\/p>\n This does not necessarily mean that OpenAI has taken over the corporate market, but it is an unwelcome sign for Anthropic. In this business, it is not falling behind that is dangerous. What is dangerous is when clients start to regard a competitor as the new technological standard. And that is precisely why the new Claude<\/a> may happen sooner than expected.<\/p>\n Reuters, citing three sources, reports that Anthropic is considering launching a new model<\/strong>, which is intended to help address OpenAI\u2019s growth following the launch of Astra.<\/p>\n However, the company has not yet announced:<\/p>\n In other words, this is not an announcement of a finished product, but rather a preview of Anthropic\u2019s next move. The very fact that such a decision has been taken is significant for the market.<\/p>\n The company\u2019s latest major release \u2014 Claude Opus 5<\/strong> \u2014 took place only on 24 July. Anthropic<\/a> stated that the model is geared towards programming and knowledge work and comes close to matching the capabilities of its most powerful systems, but is significantly cheaper to run.<\/p>\n Less than two months have passed, and the industry is already looking forward to the next generation.<\/p>\n The Astra\u2019s success should be viewed in a broader context. The main shift in the AI market at the moment is the move away from models that correspond to<\/strong>, to models that act<\/strong>.<\/p>\n Just a few years ago, the typical scenario was quite simple:<\/p>\n query \u2192 response.<\/strong><\/p>\n Now it\u2019s looking more and more like this:<\/p>\n task \u2192 search \u2192 analysis \u2192 browser \u2192 programmes \u2192 code \u2192 testing \u2192 final result.<\/strong><\/p>\n OpenAI is explicitly banking on this format. Astra demonstrates strong capabilities in computer use, web browsing, software engineering and cybersecurity. The company also describes it as the most powerful model it has widely deployed. And this is precisely the market where Anthropic has long held a very strong position. The race is no longer about, Who will write the best text?<\/strong>. The race is on to see which, who the company can entrust with the actual work<\/strong>.<\/p>\n Alongside the leap in performance, OpenAI has also seen a surge in risks. GPT-6 Astra has become the first widely deployed OpenAI model to reach the level of Critical in the field of cyber capabilities<\/strong> in accordance with the company\u2019s Preparedness Framework.<\/p>\n According to OpenAI\u2019s assessment, provided it has the appropriate tools and access, the model can:<\/p>\n OpenAI has therefore announced measures to strengthen the internal isolation of models, encrypt checkpoint files, monitor system activity and implement additional alignment checks. And this brings us to the central contradiction of the whole story.<\/p>\n Dario Amodei<\/strong> He has recently become one of the most prominent advocates of a more cautious approach to the development of the most powerful AI systems.<\/p>\n Anthropic, along with other US AI companies, warns that the capabilities of AI models are advancing faster than the mechanisms for control and regulation. European technology companies are already criticising such calls. Their argument is simple: if the US already has the most powerful models, calling on the rest of the world to \u00abslow down\u00bb can only cement American dominance. Reuters reports that this debate is now intensifying in Europe.<\/p>\n And now Anthropic finds itself in a trap. The company may believe that AI is developing too quickly. But even if it does slow down, OpenAI, Google and Meta are under no obligation to do the same.<\/p>\n Let\u2019s imagine that Anthropic delays the next version of Claude by six months.<\/p>\n During this time, OpenAI:<\/p>\n In that case, Anthropic\u2019s safer but belated release might enter a market that has already changed. This is precisely the fundamental problem with the AI race.<\/p>\n Everyone may have an interest in slowing down. But nobody wants to be the first to do so.<\/strong><\/p>\n Competitive pressure does not mean that Anthropic is in crisis. On the contrary, as of July, its The annualised revenue run rate exceeded $65 billion<\/strong>. At OpenAI, this figure stood at around $40 billion.<\/p>\n\n
Astra turned the situation around in just a few weeks<\/h2>\n
The figures that might have prompted Anthropic to act quickly<\/h2>\n
\n
What is known about the new Anthropic model?<\/h2>\n
\n
GPT-6 Astra is no longer just a chatbot<\/h2>\n
Astra has reached a critical level of cyber capabilities<\/h2>\n
\n
Anthropic is calling for a slowdown in AI development \u2014 whilst at the same time preparing a new release<\/h2>\n
Why can\u2019t Anthropic just wait?<\/h2>\n
\n
That said, Anthropic is by no means weak<\/h2>\n