Olmo

O que é Olmo?

Olmo is Ai2's family of open language models for developers and researchers. It provides model weights alongside training data, code, and evaluation tools, allowing users to examine how models are built and adapt them to their own experiments.

The current Olmo 3 family offers 7B and 32B models with Base, Instruct, and Think variants. Instruct models support multi-turn chat and tool use; Think models expose intermediate reasoning for complex prompts. You can explore models through Ai2's Playground or use official model artifacts in your own execution environment.

O que Olmo pode fazer?

Models for different tasks

Choose Base models for model development, Instruct variants for chat and tool use, or Think variants for step-by-step reasoning.

Training data access

Explore pretraining, mid-training, and post-training datasets for model development and research.

Open development tools

Use OlmoCore for training, Open Instruct for post-training, and OLMES for reproducible evaluations.

Downloadable model artifacts

Official model cards provide weights, released checkpoints, and instructions for inference and fine-tuning with Hugging Face Transformers.

Casos de uso

“Prototype a conversational application”

Use an Instruct model for multi-turn dialogue and tool-calling experiments.

“Explore reasoning behavior”

Use Think variants to investigate intermediate reasoning on complex prompts, including math and coding tasks.

“Study and adapt model training”

Work with open data, training tools, and released checkpoints for research and fine-tuning.

Perguntas frequentes

How can I try or run Olmo?

Use Ai2's Playground for chat, or download official model artifacts and run them with Hugging Face Transformers in your own environment.

Which model sizes are in Olmo 3?

Olmo 3 offers 7B and 32B models with Base, Think, and Instruct variants.

What license does the 32B Instruct model use?

Olmo-3.1-32B-Instruct is licensed under Apache 2.0. Ai2 describes the model as intended for research and educational use under its Responsible Use Guidelines.