OpenAI

SWE-bench Verified is more and more contaminated and mismeasures frontier coding progress. Our evaluation exhibits flawed exams and coaching leakage.

We share our AI mannequin’s proof makes an attempt for the First Proof math problem, testing research-grade reasoning on expert-level

OpenAI commits $7.5M to The Alignment Project to fund independent AI alignment research, strengthening world efforts to deal with AGI

OpenAI for India expands AI entry throughout the nation—constructing native infrastructure, powering enterprises, and advancing workforce abilities.

OpenAI and Paradigm introduce EVMbench, a benchmark evaluating AI brokers’ means to detect, patch, and exploit high-severity sensible contract vulnerabilities.

A new preprint reveals GPT-5.2 proposing a new system for a gluon amplitude, later formally proved and verified by OpenAI

Introducing Lockdown Mode and Elevated Risk labels in ChatGPT to assist organizations defend towards immediate injection and AI-driven knowledge exfiltration.

How OpenAI constructed a real-time access system combining rate limits, utilization monitoring, and credit to energy steady access to Sora

GABRIEL is a brand new open-source toolkit from OpenAI that makes use of GPT to show qualitative textual content and

Introducing GPT-5.3-Codex-Spark—our first real-time coding mannequin. 15x quicker era, 128k context, now in analysis preview for ChatGPT Pro customers.