Mistral Large 4
Frontier 1T-parameter MoE model in public API preview; Mistral said weights were expected at the end of October 2026, so treat it as hosted/API at this check date.
Score calculation
(4 + 2 + 4) ÷ 3 = 3.3
3 of six dimensions rated. Unrated dimensions and BDS status are excluded.
Full scoring method ↗Setup details
Release facts refer to Large 4. Company sustainability and employment policies were not verified; do not project Mistral Small 4 openness onto Large 4 before weights and license are actually published.
Mistral announced public preview of Large 4 on October 6, 2026 and said weights would drop at month end. As of October 23, the announcement supports preview availability, not verified weight publication.
Energy
↑Large 4 announcement gives size and model capabilities but no training or inference energy, water, or emissions. Sparse active-parameter counts are not an energy audit.
Reasoning & sources
The Large 4 announcement gives architecture and size information, not training energy, emissions, or inference measurements. Sparse active parameters do not prove low total energy; request measured training and serving data before comparing footprint.
No model-specific energy measurement located.
Training rights
↑the release page does not provide an itemized training corpus or rights/consent ledger for Large 4.
Reasoning & sources
Mistral’s release page does not identify the Large 4 training corpus or explain its rights process. Without those details, users cannot verify the source permissions; ask Mistral for dataset provenance.
Need full model card, datasets, license and opt-out/rights policy.
Privacy
↑Mistral's privacy policy (effective Sep. 3, 2026) covers consumer Mistral AI Products, including Mistral AI Studio, and expressly excludes business-activity processing where Mistral acts as processor. For covered consumer products, Inputs/Outputs may be used for model training subject to an account opt-out; API input/output is generally retained through generation plus 30 rolling days for abuse monitoring unless zero-data-retention is activated, while Agents API content is retained until account termination. This is company-product policy, not a Large 4 model-card property; commercial/API deployment terms may differ.
Reasoning & sources
Mistral’s current privacy policy covers consumer products and explicitly separates business processing, so the consumer terms cannot be carried over to Studio/API use. Evaluate the exact product contract and retention controls before using confidential prompts.
The consumer policy provides an account-level training opt-out and optional zero-data-retention setting, but otherwise permits training on covered Inputs/Outputs and a default 30-day API abuse-monitoring retention; score is limited to those terms.
Confirm service-specific terms and configured ZDR/retention for any particular Large 4 API deployment; policy excludes business processing and does not govern self-hosted weights.
Openness
↑Large 4 is in public preview API; the release says weights are forthcoming at end of October. It is not verified as open-weight on the check date. Mistral's Apache-licensed Small 4 is a separate release and cannot establish Large 4 terms.
Reasoning & sources
At the check date, Large 4 was in API preview and the announcement said weights would follow later in October; the record does not verify that release. At the check date, treat the checkpoint as hosted; revisit this only after confirming the weights and license are available.
As of the checked release announcement, Large 4 was API preview with weights promised later, not yet verified.
Check whether weights were actually released after October 23 and inspect the exact Large 4 license.
Worker treatment
↑the checked Large 4 announcement does not describe human data workers, annotator pay, contractor conditions, or labour protections.
Reasoning & sources
The Large 4 announcement gives no information on data workers or contractor conditions. The announcement leaves worker treatment unknown; request workforce and vendor disclosures.
No model-specific labour disclosure found.
Government & military
↑Mistral announced a November 2025 partnership with Helsing to develop vision-language-action models for defense and security applications. This documents a company/model-family defense partnership, not a Mistral Large 4 deployment or contract.
Reasoning & sources
Mistral’s Helsing announcement documents a defense/security partnership to develop specialized VLA systems, not a Large 4 deployment. It is a company-level concern, but buyers should not attribute that work to the general-purpose checkpoint without a version-specific link.
A defense partnership is disclosed at company level, but the evidence does not attribute military deployment to the Large 4 general-purpose checkpoint.
The announcement concerns development of specialized VLA systems and does not name Large 4 or disclose a complete defense customer/deployment register.
BDS status
↑We did not find Mistral AI among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.
Reasoning & sources
Mistral AI is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.
This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.