Olmo 3 7B (base)
Release in the OLMo family · version Olmo-3-1025-7B
Maintained by Ai2 (Allen Institute for AI)1
The pretrained base model at the 7B scale in Ai2's Olmo 3 release, trained in three stages (pretraining, mid-training, and long-context extension) on Dolma 3 data. The model card lists 5.93 trillion training tokens, 32 layers, and a 65,536-token context length.110
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Repository: OLMo-core Olmo 3 training scripts (external site: github.com)
- Paper: Olmo 3 technical report (arXiv 2512.13961) (external site: arxiv.org)
- Dataset hub: Pretraining data (Dolma 3 Mix used for Olmo 3 7B) (external site: huggingface.co)
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.
The model card states that the code and model are released under Apache 2.0 and that the model is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines, which list categories of prohibited use.19
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 staged intermediate checkpoints 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 lists OLMo-core as the core repository for training, inference, and fine-tuning.12 |
| Training codeIs the code used to train the model published? | Public | OLMo-core publishes the official Olmo 3 7B pretraining, mid-training, and long-context scripts.23 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The Dolma 3 Mix used for pretraining is downloadable under ODC-BY, but its dataset card says some olmOCR science PDFs were redacted after Olmo 3 7B was trained (marked [REMOVED]) and that this affects reproducibility. Mid-training and long-context mixes are linked from the model card.41 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Public | The technical report documents stages, schedules, and configurations, and the OLMo-core scripts encode the official runs.52 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Public | The model card publishes base-model evaluation results; OLMES documents commands for running the task suites from the Olmo 3 technical report.16 |
What it is useful for
Run and use notes
- The model card documents inference with Hugging Face Transformers version 4.57.0 or later, and optional 8-bit loading through bitsandbytes.1
- Intermediate checkpoints are available as Hugging Face repository revisions named stage1-stepXXX (pretraining), stage2-stepXXX (mid-training), and stage3-stepXXX (long context).1
Organization context
Provenance and derivatives
Trained from scratch by Ai2. Per the model card, stage 1 used the Dolma 3 Mix (5.93T tokens), stage 2 used the Dolma 3 Dolmino Mix (100B tokens), and stage 3 used the Dolma 3 Longmino Mix (50B tokens).14
- Derived from: Dolma 3 Mix (6T, 1025) — Stage 1 pretraining data.
- Derived from: Dolma 3 Dolmino Mix (100B, 1025) — Stage 2 mid-training data.
- Derived from: Dolma 3 Longmino Mix (50B, 1025) — Stage 3 long-context data.
Catalog records that name this entry in their provenance:
Other releases in the OLMo family
- Olmo 3 7B InstructModel-disclosure tier (USASI rubric v0.1): Open-stack
- Olmo 3.1 32B ThinkModel-disclosure tier (USASI rubric v0.1): Fully open
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.