Claude 4 models can cause the new Amnesty International in the Anthodrupry

During the opening developer conference on Thursday, Antarbur launched two new models of artificial intelligence that the startup company claims among the best in this field, at least in terms of how they are registered in popular standards.

Claude OPUS 4 and Claude Sonnet 4, part of the CLAUDE 4 family of New Clade 4 models, analyze large data sets, carry out long horizon tasks, and take complex measures, according to the company. Anthropor says that both models were seized for a good performance of programming tasks, making it perfectly suitable for writing and editing.

Both users who pay and users of free Chatbot applications for the company will be able to access the Sonnet 4, but only pay users who will get OPUS 4. For API of anthropology, via the Amazon and VERTEX AI platform from Google, the price of OPUS 4 will be at $ 15/amount of luxury (input/outputs) and Sonnet 4 For $ 15 per million dollars.

Symbols are parts of raw data in which artificial intelligence models work. One million symbols equivalent to about 750,000 words – about 163,000 words longer than “war and peace”.

Claude 4 models reach the human being, as the company looks forward to increasing revenues significantly. It is saidThe outfit, which was founded by former researchers, aims to obtain $ 12 billion in profits in 2027, an increase of $ 2.2 billion this year. man Recently closed A credit facility worth $ 2.5 billion and grew up Billion dollars From the Amazon and Other investors In anticipation of High costs Associated with the development of border models.

Competitors were not easy to maintain the pole in the artificial intelligence race. While I launched the anthropologist a A new model of artificial intelligence Earlier this year, Claude Sonnet 3.7, along with an agent coded tool called Claude Code, and competitors – including Openai and Google – to overcome the company with strong models and their Dev tools.

Play a person for Keeps with Claude 4.

Man says that Antarubor is more capable of the two models presented today, OPUS 4, can maintain a “concentrated voltage” through many steps in the workflow. Meanwhile, Sonnet 4-designer as a “alternative to drop” for SonNet 3.7-improves coding and mathematics compared to previous models of anthropology and follows the instructions more accurately, according to the company.

The Claude 4 family is also less likely than Sonnet 3.7 to engage in “piracy bonus”, as it claims Antarbur. Hacking Hacking, also known as Gaming Gaming, is a behavior where models take shortcuts and gaps to complete tasks.

To be clear, these improvements did not result in the world better Models of each standard. For example, while OPUS 4 exceeds Google’s Gemini 2.5 Pro And Openai’s O3 and GPT-4.1 On Swe-Bused, which is designed to assess the capabilities of the model coding, the O3 cannot exceed MMMU or GPQA, a set of piulia questions at the level of PhD, physics and chemistry.

However, the anthropoor launches OPUS 4 under strict guarantees, including the baccalaureate content detectors and cybersecurity defenses. The company claims that its internal tests found that OPUS 4 “may significantly increase the ability of a person who has a stem background to obtain, produce or spread chemical, biological or nuclear weapons, and it arrives Specifications of the “ASL-3” model for anthropology.

Both OPUS 4 and Sonnet 4 are “hybrid” models, as man says-capable of responses close to fixed and stretch thinking of deeper thinking (to the extent that “mind” and “thinking” as humans understand these concepts). As the thinking mode is run, models can take more time to consider possible solutions to a specific problem before answering.

It also causes models, they will show a “easy -to -use” summary of their thinking, says anthropor. Why not appear everything? Partially to protect the “competitive advantages” of anthropology, the company recognizes the draft publication blog offered to Techcrunch.

OPUS 4 and Sonnet 4 can use multiple tools, such as search engines, in parallel, and alternative between thinking and tools to improve the quality of their answers. They can also extract and save facts in “memory” to deal with tasks more reliable, and build what a person describes as “implicit knowledge” over time.

To make the models more suitable for the programmer, anthropological promotions are presented to the code of Claude mentioned above. Claude Code, which allows developers to operate specific tasks through human models directly from a station, is now integrated with IDES and provides SDK that allows Devs to be connected to third -party applications.

Claud Code SDK, which was announced earlier this week, allows Clauds Code as a subsidized operation in subsidized operation, providing a way to build auxiliary assistants and coding tools with Energy International that benefits from the capabilities of Claude models.

Antarbur has released the CLADE code extensions and connectors of the Microsoft VS, Jetbrains and GitHub symbol. GitHub connecting to developers allows for a mark on the Claude icon to respond to reference notes, as well as try to fix errors in the code – or adjust them in another way.

Artificial intelligence models are still struggling to cure quality programs. Artificial intelligence, born to a code, tends to provide security gaps and Errorsbecause of Weaknesses In areas such as the ability to understand the logic of programming. However, their promise to increase coding productivity is to push companies – and developers to Adopting them quickly.

Ethropor, completely conscious of this, is a more frequent typical update.

“We … we move on to more frequent updates, and offer a continuous flow of improvements that bring the possibilities of customers faster,” the emerging company wrote in a draft published. “This approach keeps you at the forefront and we are constantly strengthening our models and strengthening them.”