Free tool · no sign-up

Open-weight model license checker: 100 models by commercial use and OSI status

Filter 100 open-weight models by licence family, commercial use, attribution duty, restrictions and OSI approval. Every row links to the model card, which is the authoritative text.

Showing 100 of 100 models.

Snapshot 20 November 2026. “Yes*” means commercial use is allowed only if the conditions in the restrictions column are met. Model cards are the authoritative source.
#ModelLicenceCommercialAttributionRestrictionsOSI
1Llama 3.1 8B Instruct
Meta · 8B
Llama 3.1 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
2Llama 3.1 70B Instruct
Meta · 70B
Llama 3.1 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
3Llama 3.1 405B Instruct
Meta · 405B
Llama 3.1 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
4Llama 3.2 1B Instruct
Meta · 1B
Llama 3.2 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
5Llama 3.2 3B Instruct
Meta · 3B
Llama 3.2 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
6Llama 3.2 11B Vision
Meta · 11B
Llama 3.2 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
7Llama 3.2 90B Vision
Meta · 90B
Llama 3.2 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
8Llama 3.3 70B Instruct
Meta · 70B
Llama 3.3 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
9Llama 4 Scout
Meta · 109B / 17B active
Llama 4 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
10Llama 4 Maverick
Meta · 400B / 17B active
Llama 4 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
11Llama 2 7B
Meta · 7B
Llama 2 Community LicenseYes*Yes — attribution requiredAUP; 700M MAU conditionNo
12Code Llama 7B
Meta · 7B
Llama 2 Community LicenseYes*Yes — attribution requiredAUP; 700M MAU conditionNo
13Llama Guard 3 8B
Meta · 8B
Llama 3.2 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
14Mistral 7B v0.3
Mistral AI · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
15Mistral Nemo 12B
Mistral AI · 12B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
16Mistral Small 3 24B
Mistral AI · 24B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
17Mixtral 8x7B
Mistral AI · 46.7B / 12.9B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
18Mixtral 8x22B
Mistral AI · 141B / 39B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
19Codestral 22B
Mistral AI · 22B
Mistral AI Non-Production LicenseNoYes — retain licenceNon-production use onlyNo
20Mistral Large 2
Mistral AI · 123B
Mistral Research LicenseNoYes — retain licenceResearch/non-commercial; commercial licence separateNo
21Pixtral 12B
Mistral AI · 12B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
22Devstral Small 1.1
Mistral AI · 24B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
23Magistral Small 1.2
Mistral AI · 24B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
24Voxtral Small 24B
Mistral AI · 24B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
25Qwen2.5 7B Instruct
Alibaba Qwen · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
26Qwen2.5 32B Instruct
Alibaba Qwen · 32B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
27Qwen2.5 72B Instruct
Alibaba Qwen · 72B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
28Qwen2.5 3B Instruct
Alibaba Qwen · 3B
Qwen Research LicenseNoYes — retain licenceResearch and non-commercial use onlyNo
29Qwen2.5-Coder 32B
Alibaba Qwen · 32B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
30Qwen2.5-VL 7B
Alibaba Qwen · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
31Qwen3 8B
Alibaba Qwen · 8B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
32Qwen3 32B
Alibaba Qwen · 32B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
33Qwen3 235B-A22B
Alibaba Qwen · 235B / 22B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
34QwQ 32B
Alibaba Qwen · 32B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
35Qwen3-Coder 480B-A35B
Alibaba Qwen · 480B / 35B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
36DeepSeek-R1
DeepSeek · 671B / 37B active
MITYesYes — copyright noticeNone beyond the noticeYes
37DeepSeek-R1-Distill-Qwen 32B
DeepSeek · 32B
MITYesYes — copyright noticeNone beyond the noticeYes
38DeepSeek-V3
DeepSeek · 671B / 37B active
DeepSeek Model LicenseYes*Yes — retain licenceUse restrictions; commercial noticeNo
39DeepSeek-V2.5
DeepSeek · 236B / 21B active
DeepSeek Model LicenseYes*Yes — retain licenceUse restrictions; commercial noticeNo
40DeepSeek-Coder-V2
DeepSeek · 236B / 21B active
DeepSeek Model LicenseYes*Yes — retain licenceUse restrictions; commercial noticeNo
41DeepSeek-VL2
DeepSeek · 27B / 4.5B active
MITYesYes — copyright noticeNone beyond the noticeYes
42Gemma 2 9B
Google · 9B
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
43Gemma 2 27B
Google · 27B
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
44Gemma 3 4B
Google · 4B
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
45Gemma 3 27B
Google · 27B
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
46Gemma 3n E4B
Google · 8B / 4B effective
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
47CodeGemma 7B
Google · 7B
Gemma Terms of UseYes*Yes — include termsProhibited Use Policy; redistribution termsNo
48MedGemma 27B
Google · 27B
Health AI Developer Foundations TermsYes*Yes — retain termsUse restrictions; not a clinical deviceNo
49Phi-3 Mini 4K
Microsoft · 3.8B
MITYesYes — copyright noticeNone beyond the noticeYes
50Phi-3.5 Mini
Microsoft · 3.8B
MITYesYes — copyright noticeNone beyond the noticeYes
51Phi-4 14B
Microsoft · 14B
MITYesYes — copyright noticeNone beyond the noticeYes
52Phi-4 Mini
Microsoft · 3.8B
MITYesYes — copyright noticeNone beyond the noticeYes
53GPT-OSS 20B
OpenAI · 21B / 3.6B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
54GPT-OSS 120B
OpenAI · 117B / 5.1B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
55Kimi K2
Moonshot AI · 1T / 32B active
Modified MITYesYes — name display above 100M MAUAttribution above 100M MAUNo
56Kimi-VL A3B
Moonshot AI · 16B / 3B active
MITYesYes — copyright noticeNone beyond the noticeYes
57GLM-4.5
Zhipu AI · 355B / 32B active
MITYesYes — copyright noticeNone beyond the noticeYes
58GLM-4.5 Air
Zhipu AI · 106B / 12B active
MITYesYes — copyright noticeNone beyond the noticeYes
59GLM-4-9B Chat
Zhipu AI · 9B
GLM-4 LicenseYes*Yes — retain termsCommercial registration; use restrictionsNo
60GLM-4.1V 9B
Zhipu AI · 9B
MITYesYes — copyright noticeNone beyond the noticeYes
61MiniMax-M2
MiniMax · 230B / 10B active
MITYesYes — copyright noticeNone beyond the noticeYes
62MiniMax-Text-01
MiniMax · 456B / 45.9B active
MiniMax Model LicenseYes*Yes — retain termsUse restrictions; attributionNo
63MiniMax-VL-01
MiniMax · 456B / 45.9B active
MiniMax Model LicenseYes*Yes — retain termsUse restrictions; attributionNo
64MiMo 7B RL
Xiaomi · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
65Seed-OSS 36B
ByteDance · 36B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
66ERNIE 4.5 21B-A3B
Baidu · 21B / 3B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
67ERNIE 4.5 300B-A47B
Baidu · 300B / 47B active
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
68Hunyuan-Large
Tencent · 389B / 52B active
Tencent Hunyuan Community LicenseYes*Yes — retain termsExcludes EU, UK and South Korea; use restrictionsNo
69Hunyuan 7B
Tencent · 7B
Tencent Hunyuan Community LicenseYes*Yes — retain termsExcludes EU, UK and South Korea; use restrictionsNo
70Nemotron-4 15B
NVIDIA · 15B
NVIDIA Open Model LicenseYesYes — retain termsUse restrictions; indemnityNo
71Llama-3.1-Nemotron 70B
NVIDIA · 70B
Llama 3.1 Community LicenseYes*Yes — “Built with Llama”AUP; no training other LLMs; 700M MAU conditionNo
72Nemotron-H 8B
NVIDIA · 8B
NVIDIA Open Model LicenseYesYes — retain termsUse restrictions; indemnityNo
73Granite 3.1 8B
IBM · 8B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
74Granite 3.3 8B
IBM · 8B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
75Granite Code 8B
IBM · 8B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
76DBRX Instruct
Databricks · 132B / 36B active
Databricks Open Model LicenseYesYes — retain termsUse restrictionsNo
77Jamba 1.5 Mini
AI21 Labs · 52B / 12B active
Jamba Open Model LicenseYesYes — retain termsUse restrictionsNo
78Jamba 1.5 Large
AI21 Labs · 398B / 94B active
Jamba Open Model LicenseYesYes — retain termsUse restrictionsNo
79Command R 35B
Cohere · 35B
CC BY-NC 4.0NoYes — creator creditNon-commercial; no extra restrictionsNo
80Command R+ 104B
Cohere · 104B
CC BY-NC 4.0NoYes — creator creditNon-commercial; no extra restrictionsNo
81Aya Expanse 32B
Cohere · 32B
CC BY-NC 4.0NoYes — creator creditNon-commercial; no extra restrictionsNo
82Aya 23 8B
Cohere · 8B
CC BY-NC 4.0NoYes — creator creditNon-commercial; no extra restrictionsNo
83Pythia 12B
EleutherAI · 12B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
84OLMo 2 13B
AllenAI · 13B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
85OLMo 7B
AllenAI · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
86Molmo 7B
AllenAI · 7B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
87Falcon 180B
TII · 180B
Falcon-180B TII LicenseYes*Yes — attribution requiredUse restrictions; attributionNo
88Falcon 3 10B
TII · 10B
Falcon 3 TII LicenseYes*Yes — attribution requiredUse restrictionsNo
89BLOOM 176B
BigScience · 176B
BigScience BLOOM RAIL 1.0YesYes — pass on termsUse restrictionsNo
90StarCoder2 15B
BigCode · 15B
BigCode OpenRAIL-MYesYes — pass on termsUse restrictions; downstream termsNo
91StableLM 2 12B
Stability AI · 12B
Stability AI Community LicenseYes*Yes — retain termsFree under $1M revenue; enterprise licence aboveNo
92Stable Diffusion 3 Medium
Stability AI · 2B
Stability AI Community LicenseYes*Yes — retain termsFree under $1M revenue; enterprise licence aboveNo
93FLUX.1 schnell
Black Forest Labs · 12B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
94FLUX.1 dev
Black Forest Labs · 12B
FLUX.1 [dev] Non-Commercial LicenseNoYes — retain termsNon-commercial; no competing model trainingNo
95SDXL 1.0
Stability AI · 2.6B
CreativeML OpenRAIL++-MYesYes — pass on termsUse restrictions; downstream termsNo
96Whisper large-v3
OpenAI · 1.5B
MITYesYes — copyright noticeNone beyond the noticeYes
97SigLIP 2 So400m
Google · 0.9B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
98SAM 2 Hiera Large
Meta · 0.9B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
99DINOv2 Large
Meta · 0.3B
Apache-2.0YesYes — retain noticesPatent termination; state changesYes
100YOLOv8
Ultralytics · 3.2M–68M
AGPL-3.0YesYes — source and noticesCopyleft; network use triggers source offerYes

Licence families are grouped for filtering. A model card can add terms that are not visible in the licence name; open the linked card before relying on a row. This table is not legal advice.

What the checker does, and what it does not do

This page turns 100 model cards into a filterable table. It records the licence name, whether commercial use is permitted, what attribution is required, the main restriction that catches teams out, and whether the licence is approved by the Open Source Initiative. Every row links to the model card so you can read the actual text.

It does not decide whether your use is permitted. It does not cover every model in a family, every fine-tune, or every jurisdiction. It does not replace counsel. It is a triage tool: filter to the models you are considering, read the restriction column, then open the two or three model cards that matter.

Five licence buckets explain most of the 100 rows

Licence families in the snapshot. Counts are rows, not downloads or market share.
BucketModelsRepresentative licenceCommercialOSI
Apache-2.036Apache-2.0YesYes
Other custom22DeepSeek Model LicenseYes*No
Llama community14Llama 3.1 Community LicenseYes*No
MIT13MITYesYes
Non-commercial8Mistral AI Non-Production LicenseNoNo
Gemma terms6Gemma Terms of UseYes*No
AGPL-3.01AGPL-3.0YesYes

Permissive. Apache-2.0 and MIT are the simplest bucket. Commercial use, modification and redistribution are allowed, with notices and, for Apache-2.0, a patent grant and termination clause. Most Qwen, Mistral, IBM Granite, Phi and GPT-OSS releases sit here. The catch is not the licence; it is that a fine-tune may inherit another licence from its base model.

Llama community. Commercial use is allowed, but the licence adds an acceptable-use policy, an attribution phrase, a 700-million-monthly-active-user condition, and a clause that restricts using the model to improve other large language models. It is not OSI-approved. Llama 2, 3.1, 3.2, 3.3 and 4 each have their own version of the text.

Gemma terms. Gemma is distributed under a terms-of-use document rather than a conventional software licence. Commercial use is allowed subject to the Prohibited Use Policy and redistribution terms. CodeGemma, PaliGemma and MedGemma add their own conditions; MedGemma is explicitly not a clinical device.

Other custom community licences. DeepSeek, Qwen (some sizes), Falcon, Hunyuan, GLM-4, MiniMax, Jamba and DBRX use their own terms. They usually allow commercial use with restrictions, registration or scale conditions. The details differ enough that the licence name alone is not a reliable answer.

Non-commercial. CC BY-NC, the Mistral non-production and research licences, the Qwen research licence and the FLUX.1 [dev] licence do not permit commercial production use without a separate agreement. These rows are easy to spot with the commercial filter.

Commercial use is not the same as OSI approval

“Can I use it commercially?” and “Is it open source?” are different questions. A licence can allow commercial use and still fail the Open Source Definition because it restricts the field of use, requires a name display, or adds a scale condition. The OSI's definition requires free redistribution, source code, derived works, no discrimination against persons or fields of use, and no additional restrictions.

That is why Llama, Gemma and DeepSeek show “No” in the OSI column while showing “Yes*” in the commercial column. If your procurement policy says “OSI-approved licence only”, those models are out even though you can ship a commercial product with them. If your policy says “commercial use permitted”, they may be in. Write the policy in the same terms as the table.

The distinction also affects redistribution. Apache-2.0 and MIT let you redistribute the weights and derivatives under the same terms. Community model licences often require you to pass on the licence, include the acceptable-use policy, and preserve notices. A model baked into a container image still carries those duties.

Attribution and naming duties

Attribution is not one thing. Apache-2.0 and MIT require copyright and licence notices. Llama requires the phrase “Built with Llama” in the product or documentation, plus the licence. Kimi K2's modified MIT licence requires a visible name display once the product exceeds 100 million monthly active users. BLOOM's RAIL licence requires passing on use restrictions. The table records the headline duty; the model card records the exact wording.

For a product team, the practical move is to keep a licence manifest per model: base model, licence version, attribution string, notice file, and the person who approved it. Regenerate the manifest when you change the base model. A fine-tune of a Llama model is still a Llama model for licence purposes; changing the weights does not change the inherited terms.

Restrictions that catch teams out

How to check a model card in bulk

Hugging Face exposes the licence tag and card metadata through its API. This snippet prints the declared licence for a list of repositories. It reads the tag, not the full legal text, so use it to triage and then open the card.

from huggingface_hub import model_info

repos = [
    "meta-llama/Llama-3.1-8B-Instruct",
    "mistralai/Mistral-7B-v0.3",
    "google/gemma-3-27b-it",
]

for repo in repos:
    info = model_info(repo)
    licence = (info.card_data or {}).get("license", "not declared")
    print(f"{repo:45} {licence}")

Some repositories are gated: the API returns metadata but the files require accepting terms. A missing licence tag is a red flag, not permission. If the card does not declare a licence, treat the model as unlicensed until the publisher clarifies.

Methodology and limits

The snapshot was compiled on 20 November 2026 from the linked model cards and the licence texts they reference. Licence names are normalised for filtering; the original name is preserved in the table. Commercial-use, attribution, restriction and OSI fields are editorial summaries of the licence, not verbatim clauses. A model is included only if its weights are publicly downloadable at the time of the snapshot.

Three limits matter. First, a model card can be updated without changing the model name; the snapshot is a point in time. Second, a fine-tune or derivative can carry additional terms from its training data or its base model. Third, “OSI-approved” is a property of the licence text, not of the model; a publisher can apply an OSI licence to one release and a custom licence to the next. Re-check the card for the exact version you deploy.

Related reading: the Plugsky model catalogue, running models locally, and fine-tuning methods and what they inherit.

FAQ

Is an open-weight model the same as open source?

No. Open weights means the parameters are downloadable. Open source, in the OSI sense, requires a licence that permits use, modification and redistribution without field-of-use restrictions. Llama, Gemma, DeepSeek and many other model licences are not OSI-approved, even when commercial use is allowed.

Can I use Llama or Gemma models commercially?

Usually yes, with conditions. The Llama community licences allow commercial use unless the licensee had more than 700 million monthly active users at launch, and they require attribution and compliance with an acceptable-use policy. Gemma terms allow commercial use subject to the Gemma Prohibited Use Policy and redistribution terms. Read the current licence before shipping.

What does “OSI status” mean in the table?

It records whether the licence is approved by the Open Source Initiative. Apache-2.0, MIT and AGPL-3.0 are approved. Community model licences such as Llama, Gemma, Qwen and DeepSeek are not, because they add use restrictions or scale conditions that the Open Source Definition does not permit.

Do I have to display the model name in my product?

Sometimes. Llama licences require the phrase “Built with Llama” in the product or documentation, and Kimi K2's modified MIT licence requires a name display above 100 million monthly active users. Apache-2.0 and MIT require copyright and licence notices, not a product name. The table's attribution column records the duty; the linked model card is authoritative.

Is this legal advice?

No. It is a structured starting point compiled from model cards and licence texts on 20 November 2026. Licences change, model cards can be corrected, and your jurisdiction and use case matter. Have counsel review the specific licence before commercial distribution.

Sources and verification

Primary sources were opened and checked on 20 November 2026. Product terms and datasets change; the linked source is authoritative.

Need a hosted model with one less licence to review?

Plugsky serves 39 hosted models through one OpenAI-compatible API. The model catalogue lists context, modality and pricing; the free tier starts with 100 requests and 50K tokens per day.

Browse the model catalogue → See plans