Olmo 3.1 32B Think
Olmo 3.1 32B Think is an open-stack model release in the OLMo family from Ai2 (Allen Institute for AI)1, version Olmo-3.1-32B-Think, released Dec 12, 2025. Its weights, inference and training code, and training recipe are public1; the weights are under the Apache License 2.01.
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
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Repository: open-instruct Olmo 3 post-training scripts (external site: github.com)
- Repository: OLMo-core Olmo 3 training scripts (external site: github.com)
- Paper: Olmo 3 technical report (arXiv 2512.13961) (external site: arxiv.org)
On AI on Hugging Face: allenai/Olmo-3.1-32B-Think, allenai/Dolci-Think-SFT-32B, allenai/dolma3_dolmino_mix-100B-1125, allenai/Dolci-Think-RL-32B, allenai/dolma3_longmino_mix-100B-1125, allenai/Dolci-Think-DPO-32B
Availability and license
Overall availability
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.
Apache License 2.0 (external site: huggingface.co)1
Plain-language guide to the Apache License 2.0Apache License 2.0 (open-instruct) (external site: raw.githubusercontent.com)5
Plain-language guide to the Apache License 2.0Apache License 2.0 (OLMo-core) (external site: raw.githubusercontent.com)7
Plain-language guide to the Apache License 2.0
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
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Unknown — no complete fact-level rights review
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Model-disclosure tier
Open-weight, plus published inference code, training code, and training recipe, and at least documented training-data composition.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Final weights and RL step checkpoints (step revisions) are published on Hugging Face.1 |
| Inference codeIs code for running the model published? | Public | The model card documents inference with Hugging Face Transformers and vLLM.1 |
| Training codeIs the code used to train the model published? | Public | open-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 informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness. | Partial | The model card says the model was pretrained on Dolma 3 and post-trained on Dolci data and lists Dolci-Think-RL in its metadata, but its stage-by-stage section names and links the 7B datasets. The technical report says Olmo 3.1 Think 32B continued the Olmo 3 Think 32B RL run with additional epochs on the Dolci Think RL data. The 32B Dolci Think SFT, DPO, and RL dataset cards describe their composition and state that they were used for the 32B Think model, and the Dolma 3 Mix (6T) card says it was used to pretrain the 32B base model. This review confirms disclosure, not completeness across every provenance, acquisition, labeling and processing requirement.131112138 |
| Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable. | Unknown | Not assessed. |
| Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient. | Unknown | Not assessed. |
| Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above. | Public | The 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? | Public | The 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. | Public | The 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
Run and use notes
- 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
- Derived from: Olmo 3 32B (base), Olmo-3-1125-32B (external site: huggingface.co) — Base model; no separate catalog record.
- Derived from: Dolma 3 Mix (6T) — Pretraining data for the 32B base model.
- Derived from: Dolci-Think-RL-32B (external site: huggingface.co) — RL prompts used for Olmo 3 and 3.1 32B Think.
Other releases in the OLMo family
- Olmo 3 7B (base)Model-disclosure tier (USASI rubric v0.2): Open-stack
- Olmo 3 7B InstructModel-disclosure tier (USASI rubric v0.2): Open-stack
What this catalog does not know
- Training-data access: unknown.
- Complete training pipeline: unknown.
Have a primary source? How to report a correction.
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.
Share or cite this page
- X (opens X)
- Bluesky (opens Bluesky)
- LinkedIn (opens LinkedIn)
- Reddit (opens Reddit)
- Hacker News (opens Hacker News)
- Email (opens your email app)
- RSS updates
Cite this page
APA style
USASI. (2026, October 2). Olmo 3.1 32B Think. United States of America Superintelligence. https://unitedstatesofamericasuperintelligence.com/open/olmo-3-1-32b-think/
BibTeX
@misc{usasi_olmo_3_1_32b_think,
author = {{USASI}},
title = {Olmo 3.1 32B Think},
year = {2026},
month = oct,
howpublished = {\url{https://unitedstatesofamericasuperintelligence.com/open/olmo-3-1-32b-think/}},
note = {United States of America Superintelligence. Last updated 2026-10-02}
}The date is when this page was last updated. Add the date you read it if your style needs one.
For maintainers: a “Listed in USASI” badge
Links to this page. It says the entry is listed, not endorsed or rated. About the badge
Markdown (for a README)
[](https://unitedstatesofamericasuperintelligence.com/open/olmo-3-1-32b-think/)
HTML
<a href="https://unitedstatesofamericasuperintelligence.com/open/olmo-3-1-32b-think/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/listed.svg" alt="Listed in the USASI catalog" height="20"></a>