Alpha
Contact
Implement · Mistral AI

Can you run it?

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

This page is a projection of the one entry record, the Reliability and Data control factors that Implement 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

Install & run

Mistral checkpoints are standard safetensors on the verified mistralai org, and Mistral ships an official inference library. Always pin an exact revision.

# Hugging Face transformers.
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "mistralai/Mistral-Small-3.2-24B-Instruct-2506"  # Pick the exact variant to deploy.
rev = "<pinned-commit-sha>"                                    # Pin the revision; do not float on main.
tok = AutoTokenizer.from_pretrained(model_id, revision=rev)
model = AutoModelForCausalLM.from_pretrained(model_id, revision=rev)
# Official Mistral inference library.
pip install mistral-inference

# Ollama.
ollama run mistral-small

# vLLM (OpenAI-compatible server).
vllm serve mistralai/Mistral-Small-3.2-24B-Instruct-2506 --revision <sha>

# llama.cpp (community GGUF).
./llama-cli -m mistral-small-3.2.Q4_K_M.gguf -p "Hello"

Prefer safetensors, verify per-file checksums after download and confirm the model's licence.

Hardware & VRAM requirements

Mistral 7B and Nemo 12B run comfortably on a single modern GPU, and a 24 GB card handles them at full or near-full precision. Mistral Small 3.x (24B) fits a 24 GB card with modest quantization. Mixtral 8x7B needs more VRAM than its ~13B active parameters suggest, because all experts must be resident. Mixtral 8x22B and Mistral Large 2 (123B) need multi-GPU or heavy quantization for latency-sensitive serving.

Size for the largest checkpoint to be served, and budget extra headroom for the KV cache at 128K context lengths.

Serving stacks

The checkpoints are plain safetensors with standard architectures, so mainstream runtimes load them without custom code, and Mistral ships a first-party library.

StackUse it for
mistral-inferenceMistral's official reference inference
vLLMHigh-throughput serving and OpenAI-compatible API
TGIHugging Face-native production serving
llama.cppCPU, edge and GGUF quantized inference
OllamaLocal single-command runs
transformersPrototyping and fine-tuning

This breadth is a large part of why Operational Readiness scores 4. It stops short of 5 only on the narrower official quantization programme.

Safe-deployment controls & Deployment Ceiling

The model-specific gate rows from the safe-deployment playbook are below.

  • Supply chain: pin the exact mistralai revision hash and verify checksums. Prefer safetensors, which is already the default (no pickle load).
  • Model layer: deploy the instruct variant, never the base checkpoint, and set conservative decoding defaults.
  • Input and output layers: add input/output guardrails. The Mistral Moderation API (a fine-tuned Ministral 8B classifier) can serve as the companion filter, because Mistral ships no broadly distributed open guard-weight. Add prompt-injection defences and treat retrieved and tool content as untrusted.
  • Compliance: confirm the per-model licence (Apache-2.0 or research-only MRL) before commercial deployment, and archive the copyright and training-content posture in the AI-BOM.

Deployment Ceiling: T2 (customer-facing, human-reviewed), conditional. With the instruct variant, input/output guardrails, the Moderation classifier and prompt-injection defences, Mistral reaches customer-facing, human-reviewed use. It is held at T2 because safety tuning is lighter than at the largest labs and no broad independent red-team exists. T3+ (bounded autonomous) additionally needs a deterministic action-policy engine, per-action human approval for irreversible operations, and external guard tooling beyond what the release provides.

Available quantizations

A broad community ecosystem of GGUF (llama.cpp, Ollama), AWQ, GPTQ and FP8 quantizations circulates (bartowski, unsloth, lmstudio-community and others), and some official low-bit releases exist. Coverage is partial because the first-party quantization programme is narrower than the largest ecosystems and much of the breadth is community-driven. Verify the source and check quality before relying on a third-party quant in production.

Fine-tuning & adaptation

For the Apache-2.0 models, standard SFT, LoRA and QLoRA via transformers, PEFT, TRL, Axolotl or Unsloth on the safetensors checkpoints is the practical path. The specialist variants (Codestral, Mathstral, Devstral, Pixtral) show the family's adaptability.

Pre-training data and code are not published, so continued pre-training from scratch cannot be reproduced. Only the released weights can be fine-tuned. Placing a derivative on the market has EU AI Act provider implications (see Assess). An MRL model may not be fine-tuned for commercial deployment without a negotiated licence.

API / OpenAI-compatible integration

Serve Mistral behind vLLM's OpenAI-compatible /v1 server (or TGI or Ollama endpoints) for a drop-in integration with existing OpenAI-client code, or use Mistral's hosted API (Mistral Studio, formerly La Plateforme).

vllm serve mistralai/Mistral-Small-3.2-24B-Instruct-2506 --revision <sha> --port 8000
# Then point any OpenAI SDK at http://localhost:8000/v1.

No model-specific client is required for the self-hosted path. The chat-completions surface behaves like any other OpenAI-compatible endpoint, so routing, retries and gateways carry over unchanged.

Mistral's hosted API has different data terms from self-hosting. They come from the Privacy Policy (3 Sep 2026), the help centre, the Commercial Terms (25 Sep 2026) and the DPA (27 Jul 2026).

Training depends on the product. Vibe Free, Pro, Teams and Student plans train on inputs by default unless the user opts out, and Vibe Enterprise is opted out by default. Mistral Studio Free mode may train, and pay-as-you-go customers can opt out (the default is not stated). Labs and Public Preview models may always train, whatever was opted out. The Commercial Terms (4.2) allow training only on an opt-in, on feedback, as an Order Form allows, or on Labs and Preview models. DPA clause 2.3 lets Mistral train as controller unless the customer has opted out.

Retention: the API keeps input and output for the time needed to generate the output and then 30 rolling days to monitor abuse (unless ZDR is active). The Agents API keeps them until the account is terminated, and Vibe keeps them until the user deletes the conversation or the account.

Residency: hosting is in the EEA by default. An explicit US API endpoint processes requests in the US, with GDPR Article 46 safeguards (Standard Contractual Clauses) for non-EU sub-processors. The Module 4 SCCs in the DPA apply only to customers outside the EEA.

Zero Data Retention (ZDR) is available only on pay-as-you-go, for stateless API endpoints. Mistral approves it at its discretion, and it does not cover Vibe, agents, conversations, libraries, batch files or Labs and Preview models.

Self-hosting the open weights avoids all of this, because data never leaves the deployer's infrastructure.

How this scores

The ownership factors this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Moderate

The models run dependably across every major runtime. Safety tuning is lighter than at the largest US labs, so pair the instruct variant with the Moderation API and your own guardrails.

How this scores (AOI sub-dimensions)
Operational4/5how practical it is to run, serve and maintain in productionEcosystem support is broad: an official inference library (mistral-inference), day-0 or 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) and system-level guardrailing guidance, which is above a bare release.
4

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

Strong

Self-hosted, the open weights run on the deployer's own infrastructure and nothing leaves it, so data and derived knowledge stay with the deployer. Mistral's hosted services differ. Per its help centre, Vibe Free, Pro, Teams and Student plans train on inputs by default unless the user opts out (Enterprise is opted out by default), Studio Free mode may train, and Labs and Preview models may always train. Zero Data Retention exists only on pay-as-you-go stateless API endpoints. The DPA lets Mistral train as controller unless the customer has opted out (clause 2.3), applies SCCs Module 4 under French law where the customer is outside the EEA, 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.
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).