In May, Antarbur announced two new artificial intelligence systems OPUS 4 and Sonnet 4. Now, after less than six months, the company Sonnet 4.5 offers it, and it is called the best coding model in the world so far. The basis of the anthropologist for this claim is a selection of criteria in which new artificial intelligence excels not only its predecessor but also on the most expensive OPUS 4.1 systems, including Google’s Gemini 2.5 Pro and GPT-5 from Openai. For example, in Osworld, a suite that tests artificial intelligence models on computer tasks in the real world, select Sonnet 4.5 standard degrees of 61.4 percent, putting 17 percentage points above OPUS 4.1.
At the same time, the new model is able to work independently on multiple -step projects for more than 30 hours, which is a significant improvement from the seven hours or so that OPUS 4 can maintain at launch. This is an important milestone for the type of anthropitis systems that you want to build.

SonNet 4.5 outperforms the oldest models of anthropor in coding tasks and agents.
(man)
Perhaps most importantly, the company claims that Sonnet 4.5 is the safest artificial intelligence system yet, as the model has undergone “wide” safety training. This training translates into an anthropot Anthropot says it is “largely less”, deception, searching for energy and tendency to encourage fake thinking “-all the possible model features that have it Openai has fallen into hot water in recent months. Meanwhile, Antarbur has strengthened Sonnet 4.5 protection against immediate injection attacks. Due to the development of the new model, Antarbur launches Sonnet 4.5 under the AI’s safety framework of 3, which means that it comes with filters designed to prevent potential hazardous outputs related to demands about chemical, biological and nuclear weapons.

A scheme shows how Sonnet 4.5 compares to other border models in safety test.
(man)
With today’s announcement, Antarubor also introduces improvements in the quality of life through a decent Claude products. To start, Claude Code, the famous coding agent of the company, has a limited terminal, with a new feature called checkpoints. You can also guess the name, it allows you to save your progress and back to a previous case if Claude writes some unconventional software instructions that do not work completely as if you imagined it. Create the file, which is the anthropier I started going out at the beginning of the monthNow they are directly available in chats with Chatbot, and if you join the Claude for Chrome, you can start using the extension today.
API pricing for Sonnet 4.5 remains at $ 3 per million input codes and $ 15 for the same distinctive amount. Sonnet 4.5 is released from the powerful September of the Anthropor. Only one day after microsoft Claud models have been added to Copilot 365 Last week, openai I confess His rival offers the best Amnesty International for Labor Tasks.
https://s.yimg.com/ny/api/res/1.2/07mvm5dKVvVaN219Et_eYQ–/YXBwaWQ9aGlnaGxhbmRlcjt3PTEyMDA7aD02NzU-/https://d29szjachogqwa.cloudfront.net/images/user-uploaded/sonnet-4-5.jpg
Source link