# Language models - OLMo 2

OLMo 2 is a family of fully-open language models, developed start-to-finish with open and accessible training data, open-source training code, reproducible training recipes, transparent evaluations, intermediate checkpoints, and more. Details are in our [technical report](https://arxiv.org/abs/2501.00656).

## The OLMo 2 model family

### OLMo 2 32B

OLMo 2 32B is the most capable and largest model in the OLMo 2 family, scaling up the OLMo 2 training recipe used for our 7B and 13B models. It is trained up to 6T tokens and post-trained using [Tulu 3.1](/content/tulu/index.html). OLMo 2 32B is the first fully-open model to outperform GPT3.5-Turbo and GPT-4o mini on a suite of popular, multi-skill academic benchmarks.

[OLMo 2 32B release info](/content/olmo/release-notes#olmo-2-32b/index.html) [Read the blog post](/content/blog/olmo2-32b/index.html)

### OLMo 2 (7B and 13B)

OLMo 2 7B and 13B models are trained on up to 5T tokens. These models are on par with or better than equivalently-sized fully-open models, and competitive with open-weight models from Meta and Mistral on English academic benchmarks.

[OLMo 2 release info](/content/olmo/release-notes#olmo-2/index.html) [Read the blog post](/content/blog/olmo2/index.html)

### OLMo 2 1B

OLMo 2 1B is the smallest member of the OLMo model family, surpassing peer models like Gemma 3 1B or Llama 3.2 1B. The 1B model size enables rapid iteration for researchers, more local development, and a more complete picture of how our recipe scales.

[OLMo 2 1B release info](/content/olmo/release-notes#olmo-2-1b/index.html)

## The original OLMo model family

OLMo is Ai2’s first **O** pen **L** anguage **Mo** del framework, intentionally designed to advance AI through open research and to empower academics and researchers to study the science of language models collectively. More details in our [technical report](https://arxiv.org/abs/2402.00838).

[OLMo release info](/content/olmo/release-notes#olmo-february-2024/index.html) [Read the original OLMo blog post](/content/blog/hello-olmo-a-truly-open-llm-43f7e7359222/index.html)

## OLMo is fully open

Ai2 believes in the power of openness to build a future where AI is accessible to all. Open weights alone aren’t enough – true openness requires models to be trained in the open with fully open access to data, models, and code.

### Models and Data

Explore the collection of fully-open OLMo 2 models and the underlying training data used across all stages, including pre-training, mid-training and post-training –  made freely available to support open scientific research.

[OLMo 2 artifacts on Hugging Face](https://huggingface.co/collections/allenai/olmo-2-674117b93ab84e98afc72edc)

### Training

Use and extend our high-performance training code for OLMo 2, which we rely on internally for high-stakes language model training and experimentation.

[Training code](https://github.com/allenai/OLMo)

### Evaluation

Inspect the code and data used to produce OLMo 2’s results, which we make openly available for scientific reproduction and scrutiny.

[Evaluation suite](https://github.com/allenai/OLMo-Eval)

## Release notes

Get the technical details behind our OLMo model releases and subscribe to stay up to date on new releases as they happen.

[OLMo release notes](/content/olmo/release-notes/index.html)