o3, our latest reasoning model, is a breakthrough, with a step function improvement on our hardest benchmarks. we are starting safety testing & red teaming now.

An insider’s view attempting to explain to untrained mathematicians how hard FrontierMath dataset is. Average human performance given unlimited time is approximately 0%.

Excited to share our Structured Outputs sample apps repo: github.com/openai/openai-… This repo contains 3 example apps showing how to build reliable applications with LLM outputs thanks to strict JSON schemas! More details on how to use in 🧵

Logan Kilpatrick Openai swarm is great for multiagent

We're hiring Forward Deployed Engineers (FDEs)! If you're a software engineer who loves working directly with customers and helping them tap into the power of frontier AI models – DM me with an example of something cool you've built on top of our API. openai.com/careers/forwar…

openai.com/careers/forwar… I'm very pleased to announce that I'm going to lead the new Forward Deployed Engineering function here at OpenAI - our focus is getting our customers to production, whether it's through a zero-to-one novel application of our tech or helping you to scale

One of the coolest things since we launched OpenAI o1 in the API is seeing founders turn agentic workflows that weren’t reliably working—and suddenly watch them click into place with reasoning.

Please consider joining this team - a chance to become a world expert at implementing new models as well as work alongside Colin!

Over 44,000 learners have already gained valuable skills from Reasoning with o1, our latest course made in collaboration with OpenAI! In this course, Colin Jarvis, Head of AI Solutions at OpenAI, will guide you through when and how to use the o1 model for tasks requiring

My o1-based AI programming agent is now state of the art on SWE-Bench Verified! It resolves 64.6% of issues. This is the first fully o1-driven agent we know of. And we learned a ton building it.

You can now see OpenAI API pricing directly in our docs. We've made a clearer breakdown that's handy, as you're building. Copy and paste prices in table format to easily estimate costs. We've a lot more in the works here, but let me know what else we can do to improve!

we shipped a 💎 yesterday: a repo for building Realtime API voice agents with orchestration, handoffs, escalation and prompting best practices e.g. devs often under-specify the voice behavior they want. Here's a good voice prompt

We’ve put together a reference implementation for building and orchestrating agentic patterns using the Realtime API. You can use this repo to prototype a voice app using multi-agent flows in less than 20 minutes! github.com/openai/openai-…

So excited to share this with the world. It's what I wish existed when I wanted to start prototyping OpenAI realtime voice agent systems. It's really easy: clone, metaprompt for new agents, wire them up, run. You get tons of control over personality and flow behavior. Try it!

OpenAI o3-mini is fast and incredibly smart. Developers, you’ll be able to build with it in the API on day one—coming soon! 🪄

At OpenAI DevDay, we introduced the Realtime API with a demo of an interactive solar system you can navigate with your voice. A lot of you have asked how it was built, so we've open-sourced it—with a twist: this version uses our new WebRTC integration. Have fun exploring! 🌍

was so fun to collaborate with Zhiqing Sun and many many other amazing people 🙂 come work with us!

We just launched the official OpenAI Agents SDK! Inspired by our Swarm SDK, Rohan Mehta, Steven Heidel, and James Hills evolved it into an even more powerful framework that's a delight to use. We think you'll love using it! github.com/openai/openai-…

Well done Rohan Mehta Steven Heidel and James Hills, incredible ship!

Little big detail: you can pass a PDF file to the Responses API and it'll parse the file and expose the text to the LLM!