Anthropic’s new Claude AI model can decide between speed and deep thinking

Anthropic released on Monday its Claude 3.7 Sonnet model, which it says returns results faster and can show the user the “chain of thought” it follows to reach an answer. This latest model also powers a new coding tool called Claude Code that can perform some development tasks autonomously.

Claude 3.7 Sonnet offers an “extended thinking” mode that engages in a more detailed “chain of thought” reasoning but takes longer to generate a response. For simpler questions it eschews this mode and instead focuses on speed. Other models offer their own versions of “thinking” mode, but typically the user has to select that feature for harder problems;  Anthropic says Claude 3.7 Sonnet is the first publicly available model with the capability to choose the best mode based on the user’s question. If Grok 3 and DeepSeek-R1 are stick shifts, then Anthropic’s new model is an automatic.

“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,” Anthropic says in a blog post.

Claude 3.7 Sonnet outperforms other “thinking” models in some important benchmark tests. On SWE-bench, which evaluates AI models’ ability to solve real-world software issues, the model beat OpenAI’s o1 and o3-mini and DeepSeek-R1 by a comfortable margin. It was the same story on TAU-bench, which tests AI agents on complex real-world tasks with user and tool interactions. However, OpenAI’s o1 model still edges out Claude 3.7 Sonnet in math problem solving, visual reasoning, multilingual Q&A, and graduate-level reasoning benchmarks.

Anthropic describes the Claude Code tool as an active collaborator that can search and read code, edit files, write and run tests, and commit and push code to GitHub. The company says the tool has already become “indispensable” for its own coders, completing tasks in a single pass that would normally take 45 minutes or more of manual work. 

Claude 3.7 Sonnet is now available on all Claude subscription plans—Free, Pro, Team, and Enterprise–but the extended thinking mode isn’t available to users of the free tier. Claude 3.7 Sonnet is also available to developers as an API for the same price as earlier Claude models.

https://www.fastcompany.com/91283751/anthropic-new-claude-3-7-sonnet-ai-chain-of-thought?partner=rss&utm_source=rss&utm_medium=feed&utm_campaign=rss+fastcompany&utm_content=rss

Vytvorené 5mo | 24. 2. 2025, 20:20:05


Ak chcete pridať komentár, prihláste sa

Ostatné príspevky v tejto skupine

This IBM ThinkPad was astounding in 1995—and still is

Closed, it looks pretty much like any other laptop manufactured in 1995.

To be sure, it’s more compact than most—making it, in the parlance of the day, a subnotebook. But it’s still comi

15. 7. 2025, 7:20:02 | Fast company - tech
This IBM ThinkPad was astounding in 1995—and still is

Closed, it looks pretty much like any other laptop manufactured in 1995.

To be sure, it’s more compact than most—making it, in the parlance of the day, a subnotebook. But it’s still comi

15. 7. 2025, 4:50:04 | Fast company - tech
$100,000, 100 streamers: IShowSpeed and Jynxzi’s Fortnite tournament is already drawing excitement

IShowSpeed and Jynxzi are teaming up to host a $100,000 Fortnite tournament, bringing together 100 top creators for what’s shaping up to be the biggest celebrity Fortnite match to date.

14. 7. 2025, 19:40:06 | Fast company - tech
Zuckerberg announces Meta’s new AI data centers for superintelligence

Mark Zuckerberg said on Monday that Meta Platforms would spend hundreds of billions of dollars to build several massive

14. 7. 2025, 19:40:05 | Fast company - tech
Meta’s massive data center bet is a direct challenge to OpenAI and Alphabet

Meta may not currently lead the race for AI superintelligence, but it&

14. 7. 2025, 19:40:04 | Fast company - tech
Antipasto-gate: How a $40 salad sparked viral small-town drama on TikTok

Southern small-town drama has made its way to TikTok. If you’re not familiar

14. 7. 2025, 19:40:03 | Fast company - tech