USASI
Model release

Olmo 3.1 32B Think

Release in the OLMo family · version Olmo-3.1-32B-Think

Maintained by Ai2 (Allen Institute for AI)1

A 32B reasoning model from Ai2's December 2025 Olmo 3.1 update. It was produced by continuing the reinforcement learning run behind Olmo 3 32B Think for about three more weeks with additional passes over the Dolci-Think-RL data, and it writes a reasoning trace before its final answer.123

Last reviewedEntry updated Documented release Dec 12, 2025

Availability and license

Overall availability

Public

Downloadable from Hugging Face without gating. The card says the model is intended for research and educational use in line with Ai2's Responsible Use Guidelines.1

Availability is separate from permission: read the license before using or redistributing.

The model card states the model is licensed under Apache 2.0 and intended for research and educational use in accordance with Ai2's Responsible Use Guidelines, which list categories of prohibited use. The post-training datasets are ODC-BY but include third-party sources under their own licenses, per their dataset cards.11711

Model-disclosure tier

Computed from the checklist below using USASI rubric v0.1. An editorial category, not a certification.
Model-disclosure tier (USASI rubric v0.1): Fully open

Every item in the model checklist is public, including the training data itself, and the weights and code are under OSI-approved licenses.

How tiers are computed

Public materials checklist

Items for a model under USASI rubric v0.1. Unknown means unassessed or insufficient evidence.
Public materials checklist for Olmo 3.1 32B Think
ItemStatusNotes and evidence
WeightsCan the general public download the model parameters for this release?PublicFinal weights and RL step checkpoints (step revisions) are published on Hugging Face.1
Inference codeIs code for running the model published?PublicThe model card documents inference with Hugging Face Transformers and vLLM.1
Training codeIs the code used to train the model published?Publicopen-instruct publishes the Olmo 3 32B Think SFT, DPO, and RL scripts, and OLMo-core publishes the 32B base-model pretraining, mid-training, and long-context scripts.46
Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access.PublicThe 32B pretraining mix (Dolma 3 Mix 6T), the 32B mid-training and long-context mixes, and the Dolci Think SFT, DPO, and RL datasets used for the 32B Think line are downloadable from Hugging Face without gating under ODC-BY.8910111213
Training recipeAre the training configuration and procedure documented in enough detail to follow?PublicThe technical report describes Olmo 3.1 Think 32B as the Olmo 3 Think 32B RL run continued from 750 to 2,300 steps (an additional 21 days on 224 GPUs). The open-instruct README lists the 32B Think scripts and an experiment report covering 3.1.34
Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only.PublicThe model card publishes evaluation results against its SFT, DPO, and Olmo 3 32B Think predecessors; OLMES documents commands for running the Olmo 3 evaluation suites.114

What it is useful for

Ai2 describes the Think models as reasoning models aimed at math, code, and precise instruction following.151

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.
  • The model card documents inference with Hugging Face Transformers 4.57.0 or later and with vLLM, recommending temperature 0.6, top_p 0.95, and up to 32,768 generated tokens.1

Organization context

Provenance and derivatives

Trained by Ai2 from its own Olmo 3 32B base model (Olmo-3-1125-32B) through the Olmo-3-32B-Think-SFT and Olmo-3-32B-Think-DPO checkpoints, then RLVR; Olmo 3.1 extends the Olmo 3 32B Think RL run. The base model was pretrained on the Dolma 3 Mix (6T).13813

Other releases in the OLMo family

OLMo family overview

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. nonprofit or lab

The model card states the model was developed by the Allen Institute for AI (Ai2), which describes itself as a Seattle-based non-profit AI research institute.116

Assessed Sep 29, 2026

Sources

  1. 1.
  2. 2.
  3. 3.
    Olmo 3 (arXiv 2512.13961v2) (external site: arxiv.org)

    Team Olmo (Ai2) · Paper · published Apr 14, 2026 · accessed Sep 29, 2026

  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
    allenai/dolma3_mix-6T dataset card (external site: huggingface.co)

    Ai2 · Dataset card · accessed Sep 29, 2026

  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
    allenai/olmes (external site: github.com)

    Ai2 · Repository · accessed Sep 29, 2026

  15. 15.
    Olmo | Ai2 platform documentation (external site: docs.allenai.org)

    Ai2 · Documentation · accessed Sep 29, 2026

  16. 16.
    About us | Ai2 (external site: allenai.org)

    Ai2 · Official page · accessed Sep 29, 2026

  17. 17.
    Responsible use guidelines | Ai2 (external site: allenai.org)

    Ai2 · Official page · accessed Sep 29, 2026

This listing is not an endorsement, a safety assessment, or a federal approval.

Support Us

Help keep USASI useful.

Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).

About supporting this project