OLMo (Ai2): the fully open model family

Contents

1. What it does for a company
2. Strengths and limits
3. Pricing
4. Questions to ask before choosing OLMo
5. Sources

Type	Fully open model family
Headquarters	Seattle, United States
Pricing	Published
Website	huggingface.co

OLMo, built by the Seattle non-profit Ai2, is the only family reviewed here that publishes its training data and training code alongside the weights⁠12allenai/Olmo-3.1-32B-ThinkAi2 (Hugging Face)About usAi2.

Best for: Companies that need a fully open model: data, training code and weights all published.

Not the right fit for: Workloads that need frontier-scale capability.

What it does for a company

The current models are Olmo 3.1 32B, in Think and Instruct versions, and Olmo-Hybrid-7B, which has a 65,536-token context⁠13allenai/Olmo-3.1-32B-ThinkAi2 (Hugging Face)allenai/Olmo-Hybrid-7BAi2 (Hugging Face). Both are released under Apache 2.0. Ai2 publishes the Dolma 3 pretraining data and the Dolci post-training datasets, the training code (OLMo-core, open-instruct and OLMo-Eval) and intermediate checkpoints⁠13allenai/Olmo-3.1-32B-ThinkAi2 (Hugging Face)allenai/Olmo-Hybrid-7BAi2 (Hugging Face).

That makes OLMo the closest match to the Open Source Initiative’s definition of open-source AI, which asks for detailed information about the training data, the complete code used to train and run the system, and the parameters⁠4The Open Source AI Definition 1.0Open Source Initiative. For a company, the practical value is auditability: it can see what the model learned from, reproduce training and continue it with the same code on its own data. Vendor-reported, Olmo 3.1 32B Think scores 78.1 on AIME 2025 and 91.5 on HumanEvalPlus⁠1allenai/Olmo-3.1-32B-ThinkAi2 (Hugging Face).

Strengths and limits

Strengths

• The closest match to the OSI Open Source AI Definition among the families reviewed.
• Every stage of training can be inspected and reproduced.
• Small enough to fine-tune and serve on modest hardware.

Limits

• Smaller and less capable than the frontier-sized open families.
• Built by a research non-profit rather than a commercial vendor.

Choose another option if: You need the most capable model rather than the most transparent one.

Pricing

Free.

Questions to ask before choosing OLMo

• Do we need to show regulators or customers exactly what data a model was trained on?
• Is a 7B or 32B model capable enough for the work we would give it?
• Who on our team will maintain training with Ai2’s code, given that Ai2 is a research non-profit rather than a commercial vendor?
• Would we use OLMo alone, or behind a router next to a larger model?

See how OLMo compares with the other options in the main guide.

Sources

1. allenai/Olmo-3.1-32B-Think. Ai2 (Hugging Face). Accessed 9 Nov 2026.
2. About us. Ai2. Accessed 9 Nov 2026.
3. allenai/Olmo-Hybrid-7B. Ai2 (Hugging Face). Accessed 9 Nov 2026.
4. The Open Source AI Definition 1.0. Open Source Initiative. Accessed 9 Nov 2026.

Further reading

• Buyer's guide · Updated quarterlyThe best open-source AI in 2027, for companies that want to own itAn analysis of the open-weight model families a company can run today and the layer that turns one into its own, compared on seven published criteria, with sourced licences, costs and the established path from rented AI to owned AI.
• Licence guideOpen source vs open weights: what each licence lets a company doAlmost every model sold as open-source AI publishes its weights but not its training data. The difference rarely matters for internal use. It matters a great deal for companies that resell a model or build a product on it.
• Practical guideHow to train your own AI model on company dataMost of the work in training a company model happens before any training starts. A step-by-step account of the data work, the choice of base model, the training methods and the rules that apply in Europe.

Also in: Deutsch · Français · Español · Italiano · Nederlands · Polski