Anthropic unveils Claude 3.7 Sonnet, a hybrid AI model that “Thinks” on demand

Reuters

San Francisco, CA, February 24, 2025 – Anthropic has launched Claude 3.7 Sonnet, a new frontier AI model that offers users unprecedented control over its reasoning process.

Marketed as the industry’s first “hybrid AI reasoning model,” Claude 3.7 Sonnet can deliver both real-time responses and more thoroughly “thought-out” answers, with users able to decide how long the model should “think” about a given prompt.

Unlike traditional AI chatbots that require users to choose among multiple models differing in cost and capability, Anthropic’s new approach integrates reasoning into a single model. Users on Anthropic’s premium Claude chatbot plans will have access to the enhanced reasoning features, while free users will receive a standard, non-reasoning version that already outperforms its predecessor, Claude 3.5 Sonnet.

In a blog post shared with TechCrunch, Anthropic explained that Claude 3.7 Sonnet’s reasoning capability is demonstrated through a “visible scratch pad” that reveals its internal planning process - though some portions may be redacted for trust and safety. The company envisions a future where the model automatically determines how long to “think” without requiring explicit user input.

Pricing for Claude 3.7 Sonnet is set at $3 per million input tokens and $15 per million output tokens - making it more expensive than some dedicated reasoning models like OpenAI’s o3-mini and DeepSeek’s R1. However, Anthropic emphasizes that its hybrid design offers a unique blend of real-time responsiveness and deeper deductive processing.

The model has been optimized for a range of real-world tasks, including complex coding challenges and agentic tasks. In benchmark tests, Claude 3.7 Sonnet demonstrated notable improvements: scoring 62.3% accuracy on the SWE-Bench for coding tasks, compared to 49.3% for OpenAI’s o3-mini, and achieving 81.2% on TAU-Bench, a simulated retail task, outperforming OpenAI’s o1 model which scored 73.5%.

Alongside Claude 3.7 Sonnet, Anthropic is also rolling out a research preview of Claude Code, an agentic coding tool that lets developers interact with the AI directly from their terminal. Claude Code allows users to execute tasks such as analyzing project structures, making code edits, and even pushing changes to GitHub - all while providing a detailed explanation of its actions.

Anthropic’s latest release comes amid a flurry of new AI model launches from industry rivals. While the company has taken a methodical, safety-focused approach historically, its new hybrid model is a bid to lead the pack in providing versatile, high-performance AI solutions that can “think” as long as users need.

As the field continues to evolve, Anthropic hopes that integrating reasoning seamlessly with other capabilities will simplify the user experience and set a new standard for frontier AI models.

Tags

Comments (0)

What is your opinion on this topic?

Leave the first comment