The fast-moving artificial intelligence race just took another interesting turn as Anthropic — the creators of the impressive Claude — just announced a brand new version. Claude 3.7 Sonnet purports not only to be the “most intelligent” edition to date, but also the industry’s first “hybrid reasoning model”.
That sounds complex, but for customers, it’s actually really simple. It means that users looking for answers from Claude can select which “thinking mode” to use via a drop-down menu, depending on the complexity of their query.
Normal mode is “best for most use cases” — think quick fact-based queries — while the Extended option is listed as being “best for math and coding challenges”, but will likely give a more satisfying answer to anything that requires greater reasoning.
“In the standard mode, Claude 3.7 Sonnet represents an upgraded version of Claude 3.5 Sonnet,” Anthropic explains in a blog post accompanying the release. “In extended thinking mode, it self-reflects before answering, which improves its performance on math, physics, instruction-following, coding, and many other tasks.”
In extended thinking mode, it self-reflects before answering, which improves its performance on math, physics, instruction-following, coding, and many other tasks
Anthropic
This will take longer to provide an answer, of course, but we’re still talking time measured in minutes and seconds rather than hours.
You can see how this works in practice in the video below, where a user asks for an explanation of the Monty Hall problem. A quick response appears immediately, but the user then selects the extended model for the same query which not only presents a much longer answer using step-by-step thinking, but catches a mistake in its reasoning following an analysis using probability frameworks, prompting a reconsideration.
The whole thing takes 52 seconds, which is obviously longer than the quick response, but ultimately a lot more useful. The user then requests that Claude make an interactive simulator to understand the Monty Hall problem and the AI duly provides.
“We've developed Claude 3.7 Sonnet with a different philosophy from other reasoning models on the market,” Anthropic writes. “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."
Different from OpenAI
That’s in stark contrast to Open AI, which has a whole bunch of models for different needs: GPT-4, o1, o1-mini and o3-mini. While this approach offers users flexibility, it can also be confusing, and it seems the company would prefer something more streamlined. “We hate the model picker as much as you do and want to return to magic unified intelligence,” CEO Sam Altman wrote on X earlier this month.
With the new version, you might also notice a difference in how Claude reasons. "We've optimized somewhat less for math and computer science competition problems, and instead shifted focus towards real-world tasks that better reflect how businesses actually use LLMs [Large Language Models]," the company writes.
Claude 3.7 Sonnet users should see “particularly strong improvements in coding and front-end web development.”
The new version is available now, usable across all plans including the free tier. For paid users, it costs the same as the previous version: $3 per million input tokens, and $15 per million output tokens. To prevent going over budget on complex requests, API users can cap the thinking involved, by telling it not to spend more than a certain amount of tokens.
More from Tom's Guide
- New study reveals people are ditching Google for AI tools like ChatGPT search — here's why
- Help improve Tom's Guide — your chance to win a $250 Amazon gift card
- Google Gemini can now analyze and summarize documents for free — here's how
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
