Floor-weighted, not averaged. The weakest factor caps the level. Nothing here is weak, but use & modify only reaches moderate - so the substantial bar, strong on both use-and-modify and data-control, is not met, and the level is partial.

You own the Apache-licensed flagship models outright: self-host them and the model and your data are yours to run, modify and keep. Overall ownership is partial, not substantial, for one concrete reason - the family is licence-split, so a meaningful subset (Ministral 8B, Mistral Large and Pixtral Large are research-only MRL; Codestral is non-production MNPL) needs a separately negotiated commercial licence, and you cannot treat the line as uniformly free without checking each model. Training is also not transparent (data and code are closed). If you use Mistral's hosted API instead of self-hosting, turn off training and consider Zero Data Retention.

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

Moderate

The Apache-2.0 flagship models (Mistral 7B, Mixtral, Nemo, Small 3.x) can be run, modified, redistributed and used commercially with no gate - but Ministral 8B, Mistral Large and Pixtral Large are research-only MRL, and Codestral is non-production MNPL. Because the family is licence-split, you cannot treat the whole line as freely usable without checking each model, so use-and-modify is moderate rather than strong.

How this scores (AOI sub-dimensions)

Openness3/5how much is released - weights, data, code, licence - and how freelyOpen-weights tier: the flagship general-purpose models are downloadable under permissive OSI-approved Apache-2.0 - a genuinely open licence, unlike a click-through community gate - and documentation is good.

Legal4/5how permissive and clean the licence is for real commercial useA strong legal posture for the Apache-2.0 flagship models: OSI-approved, unconditional commercial use, EU-domiciled and accountable publisher, and an early signatory of the EU GPAI Code of Practice with no EU field-of-use ban.

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

Moderate

You can download and inspect the safetensors weights and read good model cards, but training data and training code are closed - you can see the model, not how it was made.

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-only distribution with per-file checksums, a clear canonical source and no malicious-checkpoint incident on record (checklist ~6/8).

Governance4/5how accountable and well-documented the publisher isReputable, legally accountable EU publisher with a predictable, dated release cadence, published model cards, a moderation offering and - distinctively - an EU GPAI Code of Practice signature.

Doesn't fail youIs it reliable and good enough for the job?

Strong

Competitive-to-strong in each size class on public leaderboards and broadly supported across runtimes; safety tuning is lighter than the largest US labs, so deploy the instruct variant plus the Moderation API.

How this scores (AOI sub-dimensions)

Performance4/5how capable it is relative to its classConsistently competitive-to-strong within each size class on public leaderboards - Mistral 7B and Mixtral were class-leading at release and Mistral Small 3.x remains strong for a 24B - and the leading European open-weight family.

Operational4/5how practical it is to run, serve and maintain in productionBroad ecosystem support: an official inference library (mistral-inference) plus day-0/near-day-0 loading on vLLM, llama.cpp, Ollama, TGI and transformers, safetensors distribution, community quants and hosted endpoints on the major clouds.

Safety3/5whether misuse risks are evaluated and guardrails are providedInstruct variants are safety-tuned and Mistral ships a hosted Moderation API (a fine-tuned Ministral 8B) plus system-level guardrailing guidance - above a bare release.

Doesn't extract your dataDoes running it keep your knowledge and data yours?

Strong

Self-hosted, the open weights run on your own infrastructure and nothing leaves - your data and derived knowledge stay yours. Mistral's HOSTED API is different: per its Privacy Policy and help centre it trains on inputs by default unless you opt out (Team/Enterprise excepted), and offers Zero Data Retention; the DPA applies SCCs Module 4 under French law and makes data inaccessible 30 days after termination.

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

Legal4/50.1612.8

Safety3/50.169.6

Performance4/50.1411.2

Operational4/50.129.6

Governance4/50.086.4

HeadlineB · 73.2/100

Dossier coverageAssess 93%Implement 96%Use 94%Support 56%How complete our four-domain documentation is, a measure of our coverage, not of the model. Each domain links to its page.

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

Documentationread2026-07-25

Mistral's model-weights documentation page, read verbatim, assigns licences per model: "Mistral 7B, Mixtral 8x7B/8x22B, Codestral Mamba, Mathstral, Mistral Nemo, Pixtral 12B, Mistral Small, Magistral Small and Devstral Small are under Apache 2 License"; "Codestral is under Mistral AI Non-Production (MNPL) License"; "Ministral 8B, Mistral Large, and Pixtral Large are under Mistral Research License."

Model cardunverified2026-07-25

Mistral open-weight checkpoints are hosted on the verified mistralai organisation on Hugging Face in safetensors with per-file checksums.

Vendor announcementunverified2026-07-25

Mistral Small 3 (24B) is released under Apache-2.0 as Mistral recommits to Apache for general-purpose models and moves away from MRL; it offers a 128K context window, tool use and structured output.

Licenceread2026-07-25

The Mistral Research Licence (MRL-0.1), read verbatim, permits use "solely for (a) personal, scientific or academic research, and (b) for non-profit and non-commercial purposes" - excluding revenue activity and SaaS distribution, under France/Paris jurisdiction; it applies to Ministral 8B, Mistral Large and Pixtral Large.

Licenceread2026-07-25

The Mistral AI Non-Production Licence (MNPL-0.1), read verbatim, restricts use: "You shall only use the Mistral Models and Derivatives for testing, research, Personal, or evaluation purposes in Non-Production Environments" - no commercial supply "including...

Vendor announcementunverified2026-07-25

Mistral Nemo (12B) is a 128K-context model released under Apache-2.0 in collaboration with NVIDIA, using the Tekken tokenizer.

Third-party analysisunverified2026-07-25

Mistral AI is a Paris-based EU company and was among the first signatories of the EU General-Purpose AI Code of Practice in 2025.

Vendor announcementunverified2026-07-25

Mistral offers a hosted Moderation API - a fine-tuned Ministral 8B classifier spanning multiple harm categories and 11 languages - plus system-level guardrailing guidance for downstream deployments.

Documentationunverified2026-07-25

Mistral documents its instruct tokenization and chat templates: mistral-common is the ground truth, newer models use the Tekken tokenizer, and the Hugging Face chat template matches mistral-common output.

Documentationunverified2026-07-25

mistral-inference is Mistral's official inference library; the models also load on vLLM, llama.cpp, Ollama, TGI and transformers.

Third-party analysisunverified2026-07-25

Mistral models are competitive-to-strong within their size classes on independent public leaderboards.

Third-party analysisunverified2026-07-25

Mistral AI is a France-based lab whose open-weight lineage (Mistral 7B, Mixtral, Nemo, Mistral Small, specialist variants) is broadly adopted, with mixed Apache-2.0 and research-only licensing and wide ecosystem/cloud availability.

Privacy Policyread2026-07-25

Mistral's Privacy Policy: by default data is hosted in the EU (servers in the EU), with a US API endpoint option that hosts data in the US; non-EU processors are covered by GDPR Article 46 Standard Contractual Clauses.

Documentationread2026-07-25

Mistral help centre: input and output data are used by default to train Mistral's models unless you opt out (via the data-sharing / 'allow your interactions to be used to train our models' control); training does not apply to Team and Enterprise plans.

Documentationread2026-07-25

Mistral help centre: for the API, input and output are kept for the period needed to generate the output and then for thirty (30) rolling days; for Le Chat, kept until you delete the conversation or your account.

Documentationread2026-07-25

Mistral help centre: Zero Data Retention (ZDR) can be activated for eligible accounts.

Data Processing Addendumread2026-07-25

Mistral's Data Processing Addendum (DPA), read verbatim, uses SCCs Module 4 (Processor-to-Controller) for transfers under French governing law (Sec 8), lists sub-processors on the Trust Centre with a 10-day objection window (Sec 7), and provides that data "will no longer be accessible upon the expiry of a thirty (30) days period following the termination" (Sec 10).