Alpha
Contact
Assess · Mistral AI

Can you own it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input B · 73.2/100

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

Mistral's open-weight family is a set of general-purpose assistant, coding and multilingual text models. It spans the dense Mistral 7B, the Mixtral 8x7B and 8x22B mixture-of-experts models, the 128K-context Mistral Nemo (12B), the 24B Mistral Small 3.x, and specialist variants (Codestral for code, Mathstral, Devstral, Pixtral for vision). It is the leading European open-weight family: EU-domiciled, strongly multilingual, and with the strongest GPAI regulatory posture in its class.

Deploy the instruct (safety-tuned) variant, not the base checkpoint, and confirm the specific model's licence before any commercial deployment. The Apache flagship models are unconditional and the MRL research-only models are not. Mistral is not positioned as a safety-critical model. 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

The main risk is the licence split. A team that assumes Mistral means Apache can trip over the research-only MRL models (Ministral 8B, Mistral Large 2, Pixtral Large) and the non-production MNPL Codestral, whose commercial use requires a negotiated licence. The split is wider than Apache versus MRL: Mistral's newest large releases (Mistral Large 4, Mistral Medium 3.5, Timestral) use custom licences with their own conditions and are not covered here.

Safety tuning is historically lighter than at the largest US labs. Instruct alignment withstands casual jailbreaks, and the companion classifier (the Moderation API) is a hosted service, not a broadly shipped open guard-weight.

Training data and code are closed, so the training pipeline cannot be inspected or reproduced and bias sources must be reasoned about from an opaque artifact. The hallucination and prompt-injection profile is standard for the class, so retrieved and tool content must be treated as untrusted.

The offsetting advantage is regulatory: an EU-domiciled, Code-of-Practice-signatory provider with an OSI-approved licence on its flagship models.

Openness tier & components

The tier is open_weights. The flagship general-purpose models ship downloadable Apache-2.0 weights under an OSI-approved permissive licence, and documentation is good. Training data and training code are closed, so there is no reproduction path. Evaluation is only partial. The licence component is split, because a meaningful subset of the family is research-only MRL.

The open-weights tier caps Dimension 1 at 3. Mistral sits at the top of that band, below the recipe (4) and fully-open (5) tiers.

License terms & permitted use

There is no single licence, so the model card decides.

The Apache-2.0 licence (OSI-approved, unconditional commercial use) covers Mistral 7B, Mixtral 8x7B and 8x22B, Mistral Nemo, Mistral Small 3.x and Magistral/Devstral Small. These can be used, modified, redistributed and commercialised, subject only to standard attribution and notice terms. Mistral has publicly recommitted to Apache-2.0 for general-purpose models with Mistral Small 3. Mistral Large 3 (675B) and Ministral 3 (3B, 8B, 14B) are also Apache-2.0, per the raw model cards (front matter plus the card's License section; those repos have no separate LICENSE file).

The Mistral Research Licence (MRL-0.1) is research-only and covers Ministral 8B (2410), Mistral Large 2 (2407, 2411) and Pixtral Large (2411). Any commercial deployment needs a separately negotiated licence from Mistral. The non-production MNPL covers Codestral-22B-v0.1, which is still tagged mnpl on Hugging Face.

The split feeds both the openness (3) and legal (4) scores. The Apache core is clean, but the family is not uniformly open.

Supply-chain & provenance

Weights are distributed from the verified mistralai org on Hugging Face as safetensors (no pickle requirement) with per-file checksums. The checkpoint trust checklist scores about 6/8. The two missing controls are cryptographic weight signing (Sigstore/model-signing) and SLSA build attestation, which is why provenance scores 4, not 5. No malicious-checkpoint incident is on record for the canonical org.

Community GGUF/AWQ/FP8 quants and hosted cloud endpoints (Bedrock, Azure AI, Vertex, Together, OpenRouter) are separate artifacts. Pin the exact mistralai revision and verify checksums on download.

EU AI Act posture

The open-weight family is a GPAI model set well under the 10²⁵-FLOPs systemic-risk threshold, so no Article 55 regime applies. The Apache releases are ≤24B dense or Mixtral MoE. Only the 123B Mistral Large 2 approaches the threshold, and it is a research-only MRL model.

The flagship Apache-2.0 models plausibly qualify for the Article 53 open-source exemption from the Annex XI/XII technical-documentation duties. The research-only MRL models carry a field-of-use restriction and do not qualify.

Mistral is EU-domiciled (Paris) and was an early signatory of the EU GPAI Code of Practice, with no EU field-of-use ban of the kind that affects some competitors. Together these make it the most EU-AI-Act-friendly non-US family in the registry. The copyright-policy and training-content-summary obligations are only partially met, because Mistral does not publish its training corpus. A fine-tuner placing a derivative on the EU market may become a provider for that derivative.

Benchmarks & evaluation

Aggregated from independent public leaderboards and third-party coverage (ev-leaderboard, ev-wiki), Mistral models are competitive to strong within each size class. Mistral 7B and Mixtral were class-leading at release, and Mistral Small 3.x remains strong for a 24B. Mistral is consistently judged the leading European open-weight family. The scores are third-party, in-class framing, so performance is capped at 4.

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

The Apache-2.0 flagship models (Mistral 7B, Mixtral, Nemo, Small 3.x) can be run, modified, redistributed and used commercially with no gate. Ministral 8B, Mistral Large 2 and Pixtral Large are research-only MRL, and Codestral is non-production MNPL. Because the family is licence-split, each model needs checking before use, so use-and-modify is moderate.

How this scores (AOI sub-dimensions)
Openness3/5how much is released - weights, data, code, licence - and how freelyThe tier is open-weights, which caps Dimension 1 at 3.
Legal4/5how permissive and clean the licence is for real commercial useThe Apache-2.0 flagship models are OSI-approved with unconditional commercial use.
2

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

Moderate

The safetensors weights can be downloaded and inspected, and the model cards are good. Training data and training code are closed, so the model is visible but how it was made is not.

How this scores (AOI sub-dimensions)
Provenance4/5how well we can trace and verify what went into the modelThe mistralai org on Hugging Face is verified.
Governance4/5how accountable and well-documented the publisher isMistral is a reputable, legally accountable EU publisher with a predictable, dated release cadence, published model cards, a moderation offering and an EU GPAI Code of Practice signature.
What this means for adoptionYou own the Apache-licensed flagship models outright: self-hosted, the model and your data are yours to run, modify and keep. Overall ownership is partial because the family is licence-split. Ministral 8B, Mistral Large 2 and Pixtral Large are research-only MRL, Codestral is non-production MNPL, and newer large releases use custom licences with their own conditions. These models need a separately negotiated commercial licence, so each model must be checked before use. Training is also not transparent, since data and code are closed. If Mistral's hosted API is used instead of self-hosting, turn off training and consider Zero Data Retention.

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.

Documentationread2026-07-25
Mistral's model-weights documentation page assigns licences per model.
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.
Licenceread2026-11-19
The Mistral Research Licence (MRL-0.1) permits use only for "Research Purposes".
Licenceread2026-11-19
The Mistral AI Non-Production Licence (MNPL-0.1) restricts use: "You shall only use the Mistral Models and Derivatives (whether or not created by Mistral AI) for testing, research, Personal, or evaluation purposes in Non-Production Environments".
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 analysisread2026-11-19
Mistral AI, a Paris-based EU company, is listed among the 22 signatories of the EU General-Purpose AI Code of Practice on the European Commission page.
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 Policyread2026-11-19
Mistral's Privacy Policy (effective 3 September 2026) covers non-EU service providers with GDPR Article 46 safeguards, with the European Commission's Standard Contractual Clauses attached.
Documentationread2026-11-19
Per the Mistral help centre, on Vibe Free, Pro, Teams and Student, input and output are used by default to train models unless the user opts out.
Privacy Policyread2026-11-19
Mistral Privacy Policy section 5 (effective 3 September 2026): for the API, Input and Output are kept for the period necessary to generate the Output and then for thirty (30) rolling days to monitor abuse (unless zero data retention is activated).
Documentationread2026-11-19
Mistral help centre: Zero Data Retention "is available only with pay-as-you-go and only for stateless API calls".
Data Processing Addendumread2026-11-19
Mistral's Data Processing Addendum (effective 27 July 2026) lets Mistral process Personal Data as Controller to train its models "unless Customer is or has opted-out of training" and for abuse monitoring except under zero data retention (2.3).
Model cardread2026-11-19
On the raw Hugging Face model cards and API, Mistral-Large-3-675B-Instruct-2512 and -Base-2512, Ministral-3-3B/8B/14B and Mistral-Small-4-119B-2603 carry license apache-2.0 in the card front matter.
Documentationread2026-11-19
The Mistral docs Models Overview labels Mistral Small 4, Ministral 3 and Voxtral Small as Apache 2.0, Mistral Large 4 as Mistral Ecosystem Open Weight License and Timestral as Modified MIT.
Terms of serviceread2026-11-19
Mistral Commercial Terms of Service (effective 25 September 2026): Mistral will not use Customer Data or Outputs to train its models except in four cases (clause 4.2).