Alpha
Contact
Assess · Ai2 OLMo

Can you own it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input B · 77.6/100

This page is a projection of the one entry record, the Use & modify and Transparency factors that Assess covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Intended & out-of-scope use

OLMo is Ai2's "open science" language-model family. It spans OLMo 2 (1B, 7B, 13B and 32B) and OLMo 3 (7B and 32B, with base, instruct and reasoning "Think" checkpoints).

Intended use is research and general-purpose assistant workloads. The differentiator is auditability: any setting where an organisation must be able to explain exactly what went into the model.

Deploy the Instruct (safety-tuned) variant. Base checkpoints are untuned research artifacts and are out of scope for customer-facing use. OLMo is not positioned as a frontier or safety-critical model, so high-stakes, autonomous or regulated decision-making is out of scope without the external control stack described in the Implement domain.

Known limitations, bias & failure modes

Capability is solid within its class and not frontier. A same-size commercial instruct model will usually edge OLMo on raw benchmarks. Instruct alignment withstands casual jailbreaks and residual risks are documented, but coverage is lighter than at the large commercial labs and no companion guard or classifier model exists.

The Dolma corpus is English-dominant, so other-language behaviour is incidental and not a supported target. Base checkpoints have no safety tuning and are research artifacts. Because the full training data is published, it can be inspected for known problem sources.

Openness tier & components

OLMo is fully_open. All six openness components are Open: downloadable Apache-2.0 weights, the full Dolma training dataset, runnable training code, released evaluation code and results, complete technical reports, and an OSI-approved licence.

This is the top "Open Science / Class I" tier of the Linux Foundation Model Openness Framework, so Dimension 1 scores 5 on third-party classification (ev-mof) and not on publisher marketing.

License terms & permitted use

Apache-2.0 covers the weights and the training code. The published data (Dolma, Dolma 3, Dolci) is ODC-By, and one Dolci subset is CC-BY-SA-4.0. Apache-2.0 is OSI-approved, its text has no field-of-use restriction, and it allows unconditional commercial use. Weights can be used, modified, redistributed and commercialised, subject only to the standard Apache attribution and notice terms.

The OLMo 3 cards and the OLMo 2 Instruct cards say the model is "intended for research and educational use in accordance with Ai2's Responsible Use Guidelines" (ev-resp-use). Those guidelines list disallowed uses outside "Research Use", including political campaigning, automated social-media posting and classifying people by protected characteristics. They are a separate Ai2 document and not part of the Apache-2.0 text, and the page does not say whether they bind someone who only downloaded the weights. Ai2's blog says the datasets carry no licence restrictions; the dataset cards say otherwise (ev-dolma3). Have counsel read the guidelines before a use case close to that list.

The caveat is scored. Legal is 4 and the use-and-modify factor is moderate, so the ownership level is partial, because substantial needs a strong use-and-modify factor. Openness (5) is unaffected, because it measures what is released and the Apache-2.0 text itself has no field-of-use limit. Ownership returns to substantial if Ai2 states that the guidelines are guidance and not a licence condition.

Supply-chain & provenance

Weights are distributed from the verified allenai org on Hugging Face in safetensors (no pickle requirement) with per-file checksums. Training data, training code and intermediate checkpoints are all public, so provenance is fully auditable and the pipeline can in principle be reconstructed.

The checkpoint trust checklist scores 6/8. The two missing controls are cryptographic weight signing (Sigstore/model-signing) and SLSA build attestation, so provenance scores 4 and not 5. No incidents or malicious-mirror findings are on record for the canonical org. Pin the exact revision and verify checksums on download.

EU AI Act posture

OLMo is a GPAI model well under the 10²⁵-FLOPs systemic-risk threshold (models are ≤32B), so no Article 55 regime applies. Released under a free/open-source licence and not monetised, it plausibly qualifies for the Article 53 open-source exemption from the Annex XI/XII technical-documentation duties.

Two obligations survive the exemption: a copyright policy and a public training-content summary. Because Dolma is fully published, OLMo can satisfy the training-content summary directly from released artifacts. This makes OLMo the most EU-AI-Act-friendly family in the registry. A fine-tuner who places a derivative on the EU market may become a provider for that derivative, but starts from a complete upstream package.

Benchmarks & evaluation

Ai2's OLMo 2/3 technical reports and independent leaderboards (ev-perf, ev-tech-report) give a consistent picture: OLMo is competitive within its size class but not frontier-leading on raw capability. A same-size commercial instruct model will usually edge it.

Its value is transparency and reproducibility, since the eval harness and results are themselves open. Treat published scores as third-party, in-class results and not as a claim of category leadership.

How this scores

The ownership factors this domain covers, drawn from the one entry record.

1

Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?

Moderate

Apache-2.0 weights, ungated, with full trainability: the OLMo-core training code and the Dolma corpus (ODC-BY) allow fine-tuning, continued pre-training or reproduction from scratch, and the licence text has no field-of-use limit. The rating is moderate because the model cards add an acceptable-use statement outside the licence: the models are 'intended for research and educational use in accordance with Ai2's Responsible Use Guidelines', and those guidelines bar uses outside Research Use. No document says whether they bind a downloader, so the strong anchor (no field-of-use restriction) cannot be asserted (ev-resp-use).

How this scores (AOI sub-dimensions)
Openness5/5how much is released - weights, data, code, licence - and how freelyAll six openness components are Open: Apache-2.0 weights, the full Dolma training dataset, runnable training code, released evaluation code and results, thorough technical reports, and an OSI-approved licence.
Legal4/5how permissive and clean the licence is for real commercial useWeights and code are OSI-approved Apache-2.0 and the published data is ODC-By.
2

TransparencyDo you know what it is: weights, training, behaviour, and legible terms?

Strong

Weights, training data, training code and evaluation are all open, so the model and exactly how it was made can be inspected. It is the most auditable model in the registry.

How this scores (AOI sub-dimensions)
Provenance4/5how well we can trace and verify what went into the modelOLMo is distributed from the verified allenai org on Hugging Face in safetensors with checksums.
Governance4/5how accountable and well-documented the publisher isAi2 is a reputable, accountable US non-profit with a clear release cadence (OLMo 2 then OLMo 3) and full technical reports.
What this means for adoptionMost, but not all, of the right to use self-hosted OLMo is owned. It is Apache-licensed, reproducible from open data and code, and self-hostable, so nothing leaves the deployer's infrastructure. The model cards tie its use to Ai2's Responsible Use Guidelines, which bar uses outside research, and no document says whether they bind a downloader. Misuse is also unbenchmarked and no companion guard model exists, so the deployer supplies the safety layer. Ownership returns to substantial if Ai2 states that the guidelines are guidance and not a licence condition. Until then, adopt OLMo where you must prove what you run, keep clear of the guidelines' disallowed uses, take advice before a use close to that list, and run the Instruct variant behind your own guardrails.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Vendor announcementunverified2026-07-25
Ai2 presents OLMo as a fully open family releasing weights, training data, code and evaluation.
Model cardread2026-07-25
OLMo 2/3 checkpoints are hosted on the verified allenai org on Hugging Face in safetensors.
Licenceread2026-07-25
The Dolma training dataset used for OLMo is publicly released under ODC-BY (Open Data Commons Attribution), transitioned to ODC-BY as of April 15, 2024, with users "also bound by any license agreements and terms of use of the original data sources."
Licenceread2026-07-25
OLMo code and weights are distributed under OSI-approved Apache-2.0.
Third-party analysisunverified2026-07-25
The Linux Foundation Model Openness Framework classifies OLMo among the most open (Open Science / Class I) model releases.
Model cardread2026-07-25
Canonical allenai repos publish per-file checksums; weights are not cryptographically signed.
Third-party analysisunverified2026-07-25
OLMo Instruct variants are safety-tuned but with lighter alignment coverage than large commercial labs.
Third-party analysisunverified2026-07-25
Independent leaderboards show OLMo competitive within its size class but not frontier-leading.
Documentationunverified2026-07-25
OLMo safetensors checkpoints load on mainstream serving stacks (vLLM, llama.cpp, Ollama, TGI, transformers).
Documentationread2026-07-25
OLMo training and evaluation code is publicly released on GitHub.
Technical_reportread2026-07-25
Ai2's OLMo technical reports, "2 OLMo 2 Furious" (arXiv 2501.00656) and "Olmo 3" (arXiv 2512.13961), document fully-released artifacts: weights, full training data, code, recipes, logs and thousands of intermediate checkpoints.
Model cardread2026-11-19
The OLMo 3 model cards (Olmo-3-1025-7B, Olmo-3-1125-32B, Olmo-3-7B-Instruct, Olmo-3.1-32B-Think) carry an Apache-2.0 tag, are not gated, and say "The code and model are released under Apache 2.0".
Terms of serviceread2026-11-19
The Ai2 Responsible Use Guidelines (page dated June 2024) "govern your use of Ai2's AI tools and artifacts, including models, datasets, APIs, and applications".
Licenceread2026-11-19
Dolma 3 (pool and mixes) is ODC-By v1.0 "intended for research and educational use" per its dataset cards.