Alpha
Ctrl K
Contact
Assess · Mistral

Can you own it?

Ownership levelLimitednone·limited·partial·substantial·fullAnalytical input C ยท 63.2/100

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

These are the Mistral models that are not Apache-2.0 - shipped under Mistral's two non-commercial licences. Under the Mistral Research Licence (MRL, research-only): Ministral 8B, Mistral Large 2 (123B), and the multimodal Pixtral Large. Under the Mistral AI Non-Production Licence (MNPL, non-production-only): Codestral. They are intended for research and evaluation. They are a separate entry from the Apache-2.0 flagship line (the mistral entry) because the licence bars commercial and production use.

Out-of-scope: any production or commercial deployment without a separately negotiated commercial licence from Mistral. For commercial work, use the Apache Mistral models or take a commercial licence.

Known limitations, bias & failure modes

The dominant limitation is the non-commercial licence - it is the reason to reach for the Apache line instead for anything beyond research. Beyond that: historically lighter safety tuning than some peers, no first-party guard model, closed training data and code, and serious infrastructure for the 123B/124B models.

Openness tier & components

open_weights tier, but with a conditional (non-commercial) licence (dimension score 2). The weights are openly downloadable and inspectable for research, but commercial/production use is barred, and training data and code are closed. That pulls it a notch below the Apache Mistral line's open-weights 3.

License terms & what you may do

Two non-commercial licences. The MRL (Ministral 8B, Mistral Large, Pixtral Large) grants use "solely for (a) personal, scientific or academic research, and (b) for non-profit and non-commercial purposes", excluding revenue activity and SaaS distribution. The MNPL (Codestral) permits only "testing, research, Personal, or evaluation purposes in Non-Production Environments", with no commercial supply including SaaS/cloud. In both cases production or commercial use requires a separately negotiated commercial licence from Mistral; outputs are not claimed by Mistral; France/Paris jurisdiction. The non-commercial bar is why use-and-modify is weak and ownership is limited.

Supply-chain provenance

The canonical source is the verified mistralai organisation on Hugging Face, with a public per-model licence table, safetensors and checksums, and no malicious-checkpoint incident (checklist ~5/8). Community quantizations circulate, but their use remains bound by the MRL/MNPL non-commercial terms - a quant confers no commercial rights. Pin the revision, verify the checksum, and confirm the per-model licence tier.

EU AI Act posture

These are GPAI models. Unlike the Apache Mistral line, the MRL/MNPL are not FOSS (they bar commercial/production use), so the Article 53 open-source exemption does not apply on the licence axis. But Mistral is EU-domiciled (France), an early GPAI Code of Practice signatory, and documents its models well, so the surviving obligations are partially met. An EU research or evaluation user is well served; a commercial deployer must negotiate a licence first. Legal scores 2.

Benchmarks & evaluation

Mistral Large 2 and Pixtral Large are strong open-weight models, and Ministral 8B is strong for its size. OneHill has not re-run these benchmarks, so performance is capped at 4 and no specific figures are asserted as verified.

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?

Weak

The MRL restricts use to research/academic/non-profit purposes, and the MNPL to non-production testing/evaluation; any commercial or production use of the model or a derivative requires a separately negotiated Mistral licence. A blanket non-commercial bar is the weakest use-and-modify position - weak.

How this scores (AOI sub-dimensions)
Openness2/5how much is released - weights, data, code, licence - and how freelyWeights are openly downloadable and inspectable, but the licence bars commercial / production use (research or non-production only), training data and code are closed, and evaluation is partial.
Legal2/5how permissive and clean the licence is for real commercial useA clear licence from an EU-domiciled provider (France law) with good documentation, but it bars commercial / production use without a separately negotiated agreement, so the open-source exemption does not apply and a commercial adopter cannot use it as released.
2

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

Moderate

Weights are inspectable and Mistral documents its models well, 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 modelVerified mistralai org on Hugging Face, safetensors with checksums, a clear canonical source and a documented per-model licence table, no malicious-checkpoint incident (checklist ~5/8).
Governance4/5how accountable and well-documented the publisher isA strong governance posture: a named, accountable, EU-domiciled provider (Mistral AI), an early GPAI Code of Practice signatory with a public licence table and good documentation.
What this means for adoptionYou have only limited ownership of these Mistral models: self-hosting keeps your data yours and the models are capable and well-supported (data-control and reliability strong), but the MRL/MNPL bar commercial and production use without a separately negotiated licence, so use-and-modify is weak - ownership is limited. For anything beyond research or evaluation, either negotiate a commercial licence with Mistral or use the Apache-2.0 Mistral models (the `mistral` entry), which carry none of these restrictions and reach substantial ownership.

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.

Licenceread2026-08-03
Mistral licence docs, read: [MRL-0.1] use "solely for (a) personal, scientific or academic research, and (b) for non-profit and non-commercial purposes", excluding revenue activity and SaaS distribution (Ministral 8B, Mistral Large, Pixtral Large).
Model cardread2026-08-03
mistralai Hugging Face org (verified) and Mistral docs: Ministral 8B, Mistral Large 2 (123B), Pixtral Large (124B multimodal) tagged MRL; Codestral tagged MNPL; safetensors with checksums; broad serving (vLLM, transformers, llama.cpp, Ollama).
Third-party analysisunverified2026-08-03
On public leaderboards Mistral Large 2 and Pixtral Large are strong open-weight models, and Ministral 8B is strong for its size; not OneHill-reproduced this session.
Third-party analysisunverified2026-08-03
Mistral has historically shipped lighter safety tuning than some peers and no first-party guard model; independent behavioural analysis is limited.
Third-party analysisunverified2026-08-03
Mistral AI is EU-domiciled (France) and an early GPAI Code of Practice signatory with public per-model licensing and good documentation, but the MRL/MNPL are non-commercial (non-FOSS), so no open-source exemption applies on the licence axis.