Model
Meta Llama 4
Publisher
Meta (United States)
Licence
Llama 4 Community License
Context
10M / 1M / unspecified
For non-EU users you partially own self-hosted Llama 4: it runs on your infrastructure, performs strongly as a multimodal family, and pairs with Meta's Llama Guard and Prompt Guard (reliability strong), but the Community Licence's 700M-MAU trigger, naming conditions, and access gate keep use-and-modify at moderate - ownership partial. For EU-domiciled individuals and companies the picture is worse: the licence denies you any grant to the multimodal models, so your effective ownership is none - prefer the Llama 3.1 text line (the meta-llama entry, no EU clause) or a non-Llama family. Either way, check the 700M-MAU trigger, honour the Llama attribution terms, and confirm your domicile against the AUP before use.
Do you really own it?
Partial
none·limited·partial·substantial·full
Analytical input: AOI C · 70.8/100
The four ownership factors
Floor-weighted, not averaged. Nothing is weak, but use & modify and data control are only moderate, so it misses the bar for substantial - strong on both use & modify and data control - and lands at partial.
1
Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?
ModerateFor non-EU users the Community Licence permits commercial use and modification below 700M MAU, with an AUP and 'Built with Llama' naming - moderate. But for EU-domiciled individuals/companies the multimodal licence is NOT granted at all, so their use-and-modify is effectively none. Scored moderate as the global position, with the EU carve-out flagged as decisive for EU entities.
How this scores (AOI sub-dimensions)
Openness3/5how much is released - weights, data, code, licence - and how freelyGated-open tier: weights are downloadable (after click-through acceptance) with strong documentation, but training data and code are closed and the licence is conditional.
Legal2/5how permissive and clean the licence is for real commercial useThe weakest legal position in the Llama family: the Community Licence is non-OSI with a 700M-MAU trigger and AUP, AND - unlike the Llama 3.x text models - denies EU-domiciled entities any licence to the multimodal models, AND the large models are systemic-risk (voiding the open-source exemption).
2
TransparencyDo you know what it is: weights, training, behaviour, and legible terms?
ModerateWeights are inspectable and documentation is strong, but training data and code are closed - open weights, closed process, so moderate.
How this scores (AOI sub-dimensions)
Provenance4/5how well we can trace and verify what went into the modelCanonical, verified (and access-gated) meta-llama org on Hugging Face distributing safetensors with checksums and per-version licence tags; no malicious-checkpoint incident on the canonical org.
Governance3/5how accountable and well-documented the publisher isA named, accountable US provider (Meta) with a verified org and a reporting path (LlamaUseReport@meta.com), meeting the score-3 anchor.
3
ReliabilityIs it reliable and good enough for the job?
StrongPerformance 4, operational 5 and safety 4 (with first-party Llama Guard / Prompt Guard): capable multimodal models with the broadest serving and a real guard stack; all at or above 4, so reliability is strong.
How this scores (AOI sub-dimensions)
Operational5/5how practical it is to run, serve and maintain in productionFirst-class ecosystem support: safetensors on every major runtime and cloud endpoint (Bedrock, Vertex, Together, Groq), extensive quantization, and the broadest toolchain of any open family.
Safety4/5whether misuse risks are evaluated and guardrails are providedA genuine strength: the family ships a first-party guard stack - Llama Guard (input/output moderation) and Prompt Guard (prompt-injection/jailbreak detection) - alongside documented safety tuning.
4
Doesn't extract your dataDoes running it keep your knowledge and data yours?
ModerateSelf-hosting keeps your data local with no telemetry or clawback, but the access gate and the EU-multimodal carve-out mean your right to use the model is conditional on domicile and licence acceptance - moderate.
How this scores
Not a scored AOI dimension. For a self-hosted model, data-control is a structural property of running the weights yourself, strong by default unless the model phones home or the licence claws back rights. For a hosted API this factor is the retention + train-on-inputs + residency read, scored from the binding terms.
How the AOI score is computed
The seven dimensions above, each scored 0 to 5, weighted and summed to the 0 to 100 headline. The score is the analytical input behind the ownership verdict, not the verdict itself.
DimensionScoreWeightPoints
Openness3/50.1810.8
Provenance4/50.1612.8
Legal2/50.166.4
Safety4/50.1612.8
Performance4/50.1411.2
Operational5/50.1212.0
Governance3/50.084.8
HeadlineC · 70.8/100
Grade ceiling: a hard flag (🚩 EU AI Act / licence - the Llama 4 Community Licence denies EU-domiciled individuals and companies any licence to the multimodal models (verbatim AUP carve-out); the large Llama 4 models also exceed 1e25 FLOP (systemic-risk), voiding the open-source exemption. Llama 3.x text models do not carry the EU carve-out.) caps the grade below the raw band. See the classification matrices.
Sources
Every rating traces to a primary document. Read means the text was verified; unverified means it is known to exist but has not yet been read.
DocumentWhat it grounds
Licenceread2026-08-03
Llama 4 Community License, read verbatim (via Meta's canonical GitHub mirror meta-llama/llama-models): the "Additional Commercial Terms" 700M-MAU threshold ("greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta"), the "Built with Llama" attribution clause, California governing law, and the incorporated AUP carrying the EU multimodal restriction.
Terms of serviceread2026-08-03
Llama 4 Acceptable Use Policy, read verbatim: "With respect to any multimodal models included in Llama 4, the rights granted under Section 1(a) ...
Model cardread2026-08-03
Llama 4 on the verified, access-gated meta-llama HF org: Scout (109B/17B, very long context), Maverick (400B/17B), Behemoth (larger); natively multimodal MoE; safetensors with checksums and a llama4 licence tag; hosted on Bedrock/Vertex/Together/Groq.
Third-party analysisunverified2026-08-03
On public leaderboards Llama 4 Maverick is competitive among open multimodal models; the family adds native multimodality and very long context.
Model cardunverified2026-08-03
Meta ships Llama Guard (input/output moderation) and Prompt Guard (prompt-injection/jailbreak detection) with the Llama family and documents safety tuning; no broad independent multimodal red-team is aggregated here.