Reasoning models like o3 are starting to aid in deep technical work and meaningful scientific discovery. Results like this will become increasingly common over the next year:

A lot of my ex-colleagues now work at OpenAI and other top-tier AI companies. When I ask them how things are going, without fail, they ALL mention how damn intense everything is. It's full-max pushing all the time. This is also borne out in my general conversations with

Best reason to switch from using 4o to o3 is that o3 agrees the best running back in the NFL is Saquon Barkley!

OpenAI o3-pro today.

I'm fortunate to be able to devote my career to researching AI and building reasoning models like o3 for the world to use. If you want to join us in pushing forward the intelligence frontier, we're hiring at OpenAI.

OpenAI Podcast Episode 2 is now live! Mark Chen and Nick Turley join @andrewmayne to pull back the curtain on the making of ChatGPT. They also get into how products are developed and what’s next for agentic coding and multimodal assistants.

I am extremely excited about the potential of chain-of-thought faithfulness & interpretability. It has significantly influenced the design of our reasoning models, starting with o1-preview. As AI systems spend more compute working e.g. on long term research problems, it is

Mark Chen AtCoder Amazing results btw. This is definitely much better than what the competitive programming community was expecting. I hope that you'll publish something about it. I'm curious about the testing budget / unique solutions created, which hopefully is something that can be disclosed.

turns out you can always just work harder

Impressive work! Makes you think about whether we’re in a simulation.

GPT-5 is here - and it’s #1 across the board. 🥇#1 in Text, WebDev, and Vision Arena 🥇#1 in Hard Prompts, Coding, Math, Creativity, Long Queries, and more Tested under the codename “summit”, GPT-5 now holds the highest Arena score to date. Huge congrats to OpenAI on this

Try GPT-5 today! One of our top level goals was to marry the pretraining and reasoning paradigms - this required lots of hard work from all of our research teams (special shoutout to Max Schwarzer) to make the models smarter, faster, more robust, and more reliable!

many plots today made me sad. this plot is important though: the models are truly getting more and more practical in the real-world

From the archive: We asked @MarkChen (Chief Research Officer at OpenAI) how chasing user metrics could harm AI development. He highlighted how chasing simple metrics like daily active users (DAU) can distort a model’s behavior over time: “If you just boost responses where

GPT-5-pro is great imo, noticeably better than o3-pro

In my opinion, the most important takeaway from this result is that our OpenAI International Math Olympiad (IMO) gold model is also our best competitive coding model. 🧵

In short 49th to 98th percentile of performance in IOI in one year without training any specialised models. Same RL as for everything else we do