OLMo 3.1 32B Think

Summary

OLMo 3.1 32B Think is a 32-billion-parameter open-weights reasoning language model from AI2's OLMo 3 family, designed to use long chain-of-thought thinking to improve reasoning on tasks like math and coding. It was pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets through SFT, DPO, and RLVR stages. AI2 releases all code, checkpoints, logs, and training details to enable open science of language models. Licensed under Apache 2.0 with an associated paper (arXiv:2512.13961).

Keywords

llm reasoning chain-of-thought open-weights olmo ai2 32b think rlvr

Topics

General / Reasoning LLM

Research notes

• Part of the OLMo 3 family (7B and 32B, Instruct and Think variants). Trained with RLVR (allenai/Dolci-Think-RL). Supports transformers 4.57.0+, vLLM, and SGLang. Uses the olmo3 architecture.