These are independent reviews of the products mentioned, but TIME receives a commission when purchases are made through affiliate links at no additional cost to the purchaser.

Eight months on from its March release, OpenAI’s GPT-4 remains the most powerful AI model to power a chatbot accessible to the public. While its predecessor, ChatGPT, performed better than just 10% of students taking the bar exam, GPT-4 outperformed 90% of them. It’s adept at verbal reasoning, can break down complicated concepts into simple language, and can even explain why a joke is funny. In September, OpenAI began rolling out the abilities to interact with the model by voice and to use images as inputs. The update, GPT-4V, was tested with Be My Eyes, an organization that builds tools for visually impaired people, and can verbally describe the contents of a picture in natural language.

Correction, Oct. 25

The original version of this story mischaracterized GPT-4. It is an AI model that powers a chatbot, not the chatbot itself.

More Must-Reads from TIME

Write to Billy Perrigo at billy.perrigo@time.com.

Hydration Made Fun
A Modern Flip Phone
A Saliva Pregnancy Test
More Than a Fast Car
A Breakthrough Bassinet
EDIT POST