Announcement_026

Today we released OLMo 3, the strongest fully-open model to-date! OLMo 3 32B Base is near Qwen 2.5 & Gemma 3 performance, and OLMo 3 32B Think is the first fully-open reasoning model to approach Qwen 3 levels. We release 12 training datasets corresponding to different staged training recipes, all open & accessible! Very proud of our team! 🥹