Why Anthropic’s latest Claude model could be the new AI to beat – and how to try it

Anthropic’s Claude AI now has a new model able to “think” longer and deeper when crafting a response to your request. 

Also: 10 key reasons AI went mainstream overnight – and what happens next

Known as Claude 3.7 Sonnet, this latest model uses advanced reasoning and greater processor time to evaluate your question in a step-by-step process and then produce a detailed result.

The new model doesn’t come free

The new extended thinking mode is accessible through the Claude website and the API for developers. But it doesn’t come free. Even though Claude 3.7 Sonnet is available for all users, you must have a Pro or Team subscription to tap into the extended thinking option.

The extended mode is particularly adept at tackling difficult math and coding problems, as well as front-end web development, Anthropic said in an announcement on Monday. In this mode, Claude “self-reflects” before it provides an answer. Taking the time to develop its response helps it better handle tasks that involve math, physics, instruction-following, and coding. Along the way, Claude shows you the steps it took to arrive at its solution.

Also: From zero to millions? How regular people are cashing in on AI

In its announcement, Anthropic also touted the performance of Claude 3.7 Sonnet in early testing and use. One site found the new model significantly improved at handling complex codebases and using advanced tools. Another successfully used Claude to build sophisticated web apps and dashboards from scratch — a task that challenged other models. In a third evaluation, Claude created production-ready code with quality designs and dramatically fewer errors.

It’s better than OpenAI’s o1?

Testing of agentic tools put Claude 3.7 Sonnet ahead of the 3.5/3.6 version of Sonnet and OpenAI’s o1 model. Testing in software engineering placed Claude 3.7 Sonnet at the top among the previous version, OpenAI’s o1, and DeepSeek R1.

Also: 3 easy side hustles OpenAI’s Operator just made possible – plus how you can get started

“We’ve developed Claude 3.7 Sonnet with a different philosophy from other reasoning models on the market,” Anthropic said in its announcement. “Just as humans use a single brain for both quick responses and deep reflection, we believe reasoning should be an integrated capability of frontier models rather than a separate model entirely. This unified approach also creates a more seamless experience for users.”

How to try Claude 3.7 Sonnet and Claude Code

You can try Claude’s new extended mode if you’re a Pro or Team subscriber. Head to the website. Click the drop-down menu for the models. Make sure Claude 3.7 Sonnet is selected and change the thinking mode from Normal to Extended. Enter and submit your request. For example, I asked Claude to create a webpage that compares the different AI models available from Anthropic, OpenAI, and Google.

Also: Will AI destroy human creativity? No – and here’s why

In response, Claude displayed each line of HTML and CSS code as it was being generated. When Claude finished its work, I was able to view the page in HTML mode or in preview mode. The only drawback here and with other types of requests is that Claude’s information is current only as of October 2024. That means it won’t be aware of real-time events and information beyond that cutoff date.

Claude 3.7 Sonnet also offers an innovation for developers. With this new version, Anthropic has introduced a command-line tool for agentic coding. Known as Claude Code, this lets developers assign intricate engineering tasks to Claude directly from the terminal.

Available for now as a limited research preview, Claude Code can search for and read code, edit files, write and run tests, and commit and push code to GitHub. Citing early testing, Anthropic said that Claude Code was able to complete tasks in a single pass that would otherwise have taken more than 45 minutes of manual labor.

Also: Anthropic offers $20,000 to whoever can jailbreak its new AI safety system

Being added to Claude Code over the next few weeks are such new features as better tool call reliability, support for long-running commands, and improved in-app rendering. Developers interested in trying Claude Code should head to the Overview site, where they can join a waitlist to sign up for the research preview.

“Our goal with Claude Code is to better understand how developers use Claude for coding to inform future model improvements,” Anthropic said. “By joining this preview, you’ll get access to the same powerful tools we use to build and improve Claude, and your feedback will directly shape its future.”