On Monday, Anthropic unveiled its new Claude 4.5 model, designed to meet the needs of businesses in the fields of coding, finance, and science. Backed by Alphabet and Amazon, the startup is seeking to differentiate itself through improved performance on long and complex software tasks. In customer tests, Claude 4.5 was able to code continuously for 30 hours, compared to 7 hours for the previous version. Internally, the Sonnet 4.5 model has also demonstrated its ability to create a complete web application.
The company is focusing on large accounts rather than the general public. The new model performs better in financial and scientific reasoning and is more efficient in using a computer, scoring 60% in operating system manipulation tests, compared to 40% for its predecessors. "Seeing the model use a computer like a person is particularly telling, especially for non-developers," said Jared Kaplan, chief scientist at Anthropic.
Anthropic is also strengthening its partnerships, with Microsoft announcing the integration of its models into the 365 Copilot suite. New tools, such as an "Agent Mode" in Excel and Word and an "Office Agent" in Copilot Chat, will be gradually rolled out. Founded by former OpenAI executives, the company emphasizes the safeguards built into Claude, which are being promoted to regulated industries. The stated goal is to prioritize reliable and consistent performance on extended tasks, rather than focusing on quick demonstrations.
