OLMo 2 32B: First fully open model to outperform GPT 3.5 and GPT 4o mini

Ai2 released OLMo 2 32B, the largest model in its OLMo 2 family, which it describes as the first fully open model to outperform GPT-3.5-Turbo and GPT-4o mini on a suite of popular multi-skill academic benchmarks. The model was trained on up to 6T tokens and post-trained with Tülu 3.1, and Ai2 says it takes one third of the training cost of Qwen 2.5 32B while reaching similar performance. All data, code, weights and details are freely available. OLMo 2 32B was trained on Augusta, a 160-node cluster provided by Google Cloud, at over 1800 tokens per second per GPU (about 38% MFU).