Screenshots from AI leaker CHOI have dropped on X today, showing a new “extended thinking” toggle and a web search tool on the Claude mobile app. There’s no information yet about how these features will work or when they’re coming, but it looks like the extended thinking feature will be for Pro users only.
Claude Extended Thinking! pic.twitter.com/DAImsEYPrt
— CHOI (@arrakis_ai) February 19, 2025
Claude is one of the lesser-known generative AI assistants out there, always losing the spotlight to ChatGPT, Gemini, and now DeepSeek as well. That doesn’t stop it from following the latest LLM trends, however. Judging by the name and the little clock icon next to it, “extended thinking” will probably be an advanced reasoning feature that will let Claude spend longer thinking about its answers and show its thought process.
This could be similar to OpenAI’s o1 and o1-mini models, which “think” for longer and include some “reasoning” in their answers. It could also be similar to DeepSeek R1’s chain-of-thought feature that includes a step-by-step explanation of the model’s thought process separate to its final answer. Gemini’s Flash Thinking model also shows a detailed breakdown of the “thinking process” used to reach an answer or generate a response.
The main difference between the advanced reasoning features of these competing models is how much of the thought process we see. OpenAI, for example, chose to hide the chain of thought of its o1 model so it could be left completely unfiltered and used for monitoring purposes. No one has any idea what kind of content will pop up in a completely unedited version of an LLM’s “thought process,” so the company understandably doesn’t want to show it to users.
By that logic, the chain-of-thought explanations that DeepSeek R1 and Gemini Flash Thinking show are likely edited or controlled in some way to prevent too many strange “thoughts” popping up. We don’t know which way yet Claude will lean with its extended thinking feature but the ultimate goal is likely the same — to take more time and produce more “reasoned” responses to complex prompts.
As for the web search beta that shows on the second screenshot, directly accessing the web is something Claude users have been wanting for a while. Until now, the model has been designed to be “self-contained,” but with so many other companies adding web-surfing abilities to their models, it seems Anthropic has deemed it a necessary addition to stay competitive.
It’s unclear right now who will get access to these new features and when, but I hope we’ll all get the chance to play around with them at least a little bit.
