As the AI models gain a leap in making responses quicker and more efficient, Anthropic unveils its latest AI model, Claude 3.7, equipped with an all-new “Extended Thinking Mode,” enabling the developers to customize the thinking time that the AI model takes to generate a response. This helps the AI model to calculate the response time and apply less or more cognitive ability based on the amount of deliberation a task needs.

Enabling the mode allows the AI model to dedicate more time to thinking and expend more effort in generating a response. However, the mode change doesn't mean switching to a different model with a different strategy. Lastly, the all-new feature aims to make the AI model more intelligent and smart in making decisions regarding spending its time and effort.

Claude 3.7: Visible Thought Process 

The all-new Claude 3.7 comes with yet another addition, the "Visible Thought Process," which enables the users to trace the model's approach to any query in raw form. This means you can now see how Claude thinks, which will further allow you to get better outputs by tracking down the loopholes in the model’s approach.

Finally, it will help you identify cases when an AI model is involved in deception, behaving opposite to what it thinks, also termed as fake alignment.

Read about the latest updates in Grok 3 at: Elon Musk’s XAI Upgrades Grok 3 to Talk to You in Real-time Voice Mode - See, How It Works

Claude 3.7: More Computational Power 

As an agent, it develops a knack as it spends some time with any activity, like playing games on its own or similar tasks. It can allocate more turns, time, and computational power to computer-use tasks as an agent. The company even cites the test results of the model playing “The Pokemon Go” game, where it successfully battles three Pokémon Gym Leaders (the game’s bosses) and wins their Badges.

The model is so capable as an agent because of its ability to try multiple strategies simultaneously and question its previous assumptions, which further allows it to improve its capabilities as it progresses.

Claude 3.7: Preventing Unintended Actions

The model is now more efficient, up to 88 percent, in preventing malicious third-party hidden messages planted on the computer, tricking it into unintended actions. It is up from 74% of the time without mitigations.

Register to attend our upcoming show at: Experience Tech Firsts & Meet boAt’s Aman Gupta at Mobile India Expo 2025 – Register Now!