The latest model AI Antropic spent 30 hours to encode a chat -like application to Slack or bands. According to Anthrop, he spit out about 11,000 code lines, and stopped working only when he completed the task.
The model, Claude Sonnet 4.5, has been announced today, and its ability to act autonomically for 30 hours is a huge leap forward. Earlier, the OPUS 4 model was in May because its ability to operate for seven hours.
This is all a significant step in the battle of Anthropic to turn off the market of both AI agents and AI coding. A company called Claude Sonnet 4.5 “The best model in the world for real agents, coding and using a computer” and said that “it runs the market in using computers”, referring to the function of the computer, Antropic debuted Almost a year ago. According to Anthropic, the fresh model is particularly expert in areas such as cyber security, financial services and research. One of his Beta, Canva tests, said that the fresh model helped in “complex tasks with a long context-from engineering in our code database for product and research features.”
Anthropic, Opeli, Google and other companies constantly publish incremental updates and functions that allow their technology to act as an assistant for both consumers (studying topics, planning meetings and browsing flights) and in the case of using enterprises and programmers (creating decks of the slides, helping in coding and analyzing spreadsheets). The battle for attention and reaction warms up almost every month, if not every week. A few days ago, Opeli announced Pulse, his latest chatgpt function designed as part of the morning routine of users and research topics related to their days.
Anthropic also said that the fresh model will be combined with other updates to support programmers coding their own AI agents.
“We combine the launch of the model with access to virtual machines, memory, context management and support of many agents,” the company in the edition invited. “This basically packs the same structural elements that fed the Claude code – enabling programmers to build their own latest agents.”
Dianne Penn, head of product management in Anthropic, said The Verge In an interview that the model improvements in the possibilities of using a computer surprised her even. Claude Sonnet 4.5 is more than three times qualified in navigation in the browser and using the computer compared to the Anthropic technology of October last year. Penn said that the team received feedback from early access customers-“Githubs and the World Cursors”-and spent the last month working intensively on the model.
Scott White, product cable for Claude.Ai, he said The Verge The fact that the fresh model works at the “level of the chief of staff” and can find accessibility between the calendars of many people and plan a meeting, look at the data desktop and combine observations, write status updates based on one -on -one meetings one of its direct reports and more.
Neither White nor Penn have yet tried to encode the atmosphere with a fresh model when The Verge I talked to them. But Penn said that she is using Claude Sonnet 4.5 to employ potential fresh team members in Anthropic.
“It was really very helpful that I use:” Make a deep search on the web, invent these parameters for profiles that will allow some types of roles in my team, “said Penn. “It was really very helpful. And I saw Sonnet 4.5 coping with even better than in the past, about the quality and deaf of the search and actually generating a spreadsheet with the LinkedIn profile, so I can send them e -mail.”