Most of the frontier AI chatbots available commercially right now can offer models that think through their tasks. That is, they can take a bit longer to reason before delivering the answer.
ChatGPT o3 is one such example, and it’s my new default ChatGPT model. I like the idea of an AI model taking some time to figure out the answer rather than spitting out information the second I submit the prompt. Even then, o3 doesn’t always give me what I want, so I often have to go back and forth with it.
That reasoning can be useful for various prompts, not just simple questions that usually require a few online searches, but also more complex tasks like coding or problem-solving.
It’s not just ChatGPT that has reasoning. You’ve got access to a variety of AI options to test, so you’re not limited to OpenAI. DeepSeek is a well-known ChatGPT alternative that surprised the world in early 2025 with a reasoning model on par with ChatGPT o1. Gemini and Claude are two more examples.
If you prefer the latter, you’ll be glad to hear that Anthropic is about to give Claude a big upgrade in reasoning abilities. The AI will be able to return to reasoning to help you out, which might include testing code and fixing errors along the way.
According to The Information, via Tibor Blaho, Anthropic is updating both Claude Sonnet and Claude Opus with the new thinking abilities.
The two updates should roll out in the coming weeks. Once they’re live, Claude Sonnet and Opus will be able to switch between various tools and apps during the thinking phase to find the right answer.
Even better, if the AI gets stuck trying to solve a problem, it’ll go back to reasoning to figure out what’s wrong and try to self-correct. The feature will work for a range of prompts that Claude can handle, including code generation.
Apparently, Claude will be able to test the code it generates and spot mistakes. If it finds any, it’ll pause to consider what caused the error and attempt to fix it.
It’s unclear if any of that will be visible to users or if you’ll just get the finished response after the reasoning rounds are done.
A chat with ChatGPT o3, one of OpenAI’s most advanced reasoning models available to ChatGPT users, told me that the features Claude is about to get are already available in the more advanced OpenAI models.
ChatGPT o3 can use tools during the thinking process, like browsing the web or writing code when the task requires it. The AI can return to reasoning and use additional tools if needed.
Tapping on the Thinking icon in ChatGPT o3 and other reasoning models shows you that process. You’ll see the AI’s thoughts and how many times it might have performed an online search while figuring out the answer to your prompt.
Some of that reasoning might happen under the hood. We won’t get access to all of it, o3 claims, for various reasons. That’s why we might not always be aware of how the AI operates while thinking about a prompt. We may just see the final answer the AI thinks is best for us.
Back to the rumored reasoning upgrade for Claude, I’d expect users to notice better performance and experience once Anthropic rolls it out. We’ll just have to wait for the company to explain how it enhanced Claude’s reasoning and share benchmarks to back it up.
