Anthropic’s New Model Excels at Reasoning and Planning—and Has the Pokémon Skills to Prove It

Claudie 3.7 played the Sonnet game, fled to some difficulties: “dozens of hours“After watching the long-term memory and planning capacity of Claude, which continued to play with the long-term memory and planning opportunities in a city and continued to play with the progress in the game, Hershey shows a new compatibility level, meaning a new fitting level.

“This is one of my favorite ways to get acquainted with a model. Like, I understand what his strong parties are,” says Hershey. “This is my way of meeting with this new model and how to work with us.”

Everyone wants an agent

Anthropik’s Pokémon Study is a novel approach to solve a pre-prepared problem – we realize how the EU understands how the EU understands and the correct direction?

The answer to this question is inseparable to develop a very rejoicing AI agents of the industry that can solve complex tasks with complex duties in the industry. In Pokémon, it is important that the model does not lose context or to “forget” the task in hand. EU wanted to automate a workflow that took hundreds of hours to agents.

“A task is to be a five-minute task for a 30-minute task, you can see everything you need to keep in the sequence of the model, to remember everything you need to do [the task] He successfully worsens over time, “Hershey said.

Anthropic, Like many other AI labsHopes to create strong agents to sell as a product for consumers. Krieger says the anthropick’s “best goal” This year “hours of working hours for you” Claude.

“Now in this model, this model, the first entry is a great refrain from our customers and is often touching on the process of reconstruction of the code, often referring to the process of more efficient and organizing.

This is the future that companies are facing like Google and Openai. This week ago, Google released Mariner, An AI agent in chrome These groceries (per month) can do tasks such as $ 249.99). Recently openai Left the coding agentand back a few months Launched the operatorAn agent who can view the internet on behalf of a user.

Compared to competitors, the anthropic often seems to be a more careful mover over the research, but it looks slower in placement. And with strong AI, it is probably positive: There are many things that can go wrong with an agent who obtains sensitive information such as the inbox or bank entries. Anthropic in a blog post on Thursday: “We will be significantly reduced in places where models use shortcuts or gaps to perform tasks.” The company is also less than 65 percent in the same behavior in both Opus 4 Opus and Claus, and this behavior is less likely to deal with this behavior, known as award hacker, for at least certain coding tasks.