My life has been enriched by joyful learning. Few things give me greater joy than sharing my love of learning with others. Which is why today is a momentous day. With gratitude to my Creator, Source of Life, for allowing me to reach this day, I'm announcing that we at

OpenAI just cannot stop winning

Token crisis: solved. ✅ We pre-trained diffusion language models (DLMs) vs. autoregressive (AR) models from scratch — up to 8B params, 480B tokens, 480 epochs. Findings: > DLMs beat AR when tokens are limited, with >3× data potential. > A 1B DLM trained on just 1B tokens

Presenting the GLM-4.5 technical report!👇 arxiv.org/abs/2508.06471 This work demonstrates how we developed models that excel at reasoning, coding, and agentic tasks through a unique, multi-stage training paradigm. Key innovations include expert model iteration with

🎯 Check out Marco-Voice: A Unified Framework for Expressive Speech Synthesis with Voice Cloning 🎧 Key Features: 🔥 Novel Methods: speaker-emotion disentanglement and rotational emotion embedding integration 🔥 New Benchmark: high-quality emotional speech dataset (10 hours, 7

We are excited to officially release RynnVLA-001, a new open-source Vision-Language-Action model! 🤖 Our model outperforms strong baselines like Pi-0 & GR00T-N1.5 in real-world robot manipulations. This is achieved through several key innovations: 🔹 Generative Pre-training:

Can AIs beat long video games? We made TextQuests to test GPT-5, Grok 4, Deepseek, etc. These games can often take people dozens of hours to beat. - AIs can't beat any of the games (without clues) - some AIs behave more viciously than others - AIs are getting better rapidly

We are super excited to release OpenCUA — the first from 0 to 1 computer-use agent foundation model framework and open-source SOTA model OpenCUA-32B, matching top proprietary models on OSWorld-Verified, with full infrastructure and data. 🔗 [Paper] arxiv.org/abs/2508.09123 📌

Used to sit all day on my MacBook. Tried a standing desk—hated it for not moving all day. Now I voice-input on an 8" tablet while walking around a mall and outside and stop at random spots. Feels like going back to the time when people spent most of day walking and standing.

Being Japanese is funny sometimes. Every few years, America “discovers” something I grew up with. Suddenly I’m sophisticated, just for existing. Matcha ice cream? Been eating it since day one. Welcome to the party.

Today, we're bringing agentic capabilities to AI Mode in Search for Google AI Ultra subscribers. But... what is actually different? Let's say you want to make a dinner reservation. Traditionally, that would require multiple searches, concurrent tabs, and a lot of manual

Introducing NEO: The first Autonomous Machine Learning Engineer. It works like a full-stack ML engineer that never sleeps: handling data exploration, feature engineering, training, tuning, deployment, and monitoring, end to end. Powered by 11 specialized agents, NEO runs

Our models run real time on a laptop 5090 at 100+ FPS. We're so excited for people to start playing with it.

Router wasn't learning at first, we debugged it step-by-step and showed you how despite perfect load balancing, routing can be completely useless. We root caused it and fixed the problem. Papers skip the methodology, but you can find all details in our part 3 of MoE 101 series