AIvengo > Reviews > DeepSeek R1 after minor update approached OpenAI O3 level

DeepSeek R1 after minor update approached OpenAI O3 level

DeepSeek company released an update for its R1 language model. Despite developers calling it “minor”, many users note substantial improvements. They report that the model’s reasoning has become noticeably deeper and longer, approaching the level of O3 from OpenAI – one of the most powerful models on the market.

Simply no words! Screenshots have even appeared online where the model demonstrates continuous reasoning for longer than 20 minutes. Well, this is a significant performance leap. The update announcement was made through the company’s official account on Chinese social network WeChat, and you can test the new capabilities right now on the official website deepseek.com.

There are no official performance tests yet that could confirm user claims about reaching O3 level. But the expert community is already eagerly awaiting benchmark publications to get an assessment of the new update. If rumors are confirmed and DeepSeek R1 has indeed achieved performance comparable to OpenAI’s O3, this is really cool. I think that if R1 after a simple update approached O3 level, then the future DeepSeek R2 version may open an era of truly “thinking” systems capable of maintaining deep analysis at a level comparable to human experts. Exciting and alarming.

Autor: AIvengo

For 5 years I have been working with machine learning and artificial intelligence. And this field never ceases to amaze, inspire and interest me.

USA introduced first robot police officer based on Ford Explorer

The USA introduced the first robot patrol police officer, and you know what? It looks like a regular car, only packed with AI to the brim. This unmanned patrol vehicle is called Police Unmanned Ground Vehicle, or simply PUG.

Startup Enveda uses AI to create drugs from ancient herbs

A biotech startup from Colorado combined ancient wisdom with artificial intelligence to create new drugs. Company Enveda developed a platform that describes molecular formulas of medicinal herbs and predicts their therapeutic potential.

Channel 4 conducted experiment with AI presenter

British TV channel Channel 4 conducted a real psychological experiment for viewers. It showed an entire program about AI's impact on the job market that was hosted by a generated presenter. And only announced this at the very end. Goal? To show how easy it is to be deceived by content that's hard to verify.

19% of high schoolers in USA had romantic relationships with chatbots

Researchers in the USA found something interesting in a new report on AI's impact on school life. Turns out a whole 19% of high schoolers have already had romantic relationships with a chatbot or know someone who has. Think about that number.

Cameras on workers' heads train AI to understand physical labor

Startup Turing Labs figured out how to train AI to see and understand the world. Strap GoPro cameras on freelancers' heads and make them work. Sounds like mockery? Perhaps. But the method seems to work.