GPAI Ledger › Ministral 3 14B (Mistral AI) › Capture 11 Aug 2026

Ministral 3 14B — capture 20260811T104432Z

Verify: sha256sum 4198542c9ac7776856d379c957c323113950408dd0a72ecaa0095f7ed0c1f28d.pdf must equal the hash above (the filename IS the expected hash); ots verify 4198542c9ac7776856d379c957c323113950408dd0a72ecaa0095f7ed0c1f28d.pdf.ots -f 4198542c9ac7776856d379c957c323113950408dd0a72ecaa0095f7ed0c1f28d.pdf (opentimestamps.org) proves the capture time (fresh proofs report 'pending' until bitcoin-anchored, typically within a day).

Extracted text

Machine-extracted text (layout may be lost; the authoritative content is the stored file above).

Public  Summary  of  Training  Content   Version  of  the  Summary:   v.1  Last  update:   31/07/2025  General  information
1.  General  information  1.1.  Provider  identification
Provider  name  and  contact  details:
Mistral  AI  15  rue  des  Halles,  75001  Paris  FRANCE
Authorised  representative  name  and  contact  details:
NA.
 1.2.  Model  identification
Versioned  model  name(s):
 This  Public  Summary  of  Training  Content  applies  to  the  following  versions  of  Ministral  3  -14B:  Base,  Instruct  and  Reasoning.
Model  dependencies:
 Mistral  Small  3.1,  released  on  March  17,  2025   Date  of  placement  of  the  model  on  the  Union  market:
December  2,  2025  (applicable  to  Base,  Instruct  and  Reasoning).
1.3  Modalities,  overall  training  data  size  and  other  characteristics   Modality  Training  data  size
Types  of  content
 ☒  Text
☐  Less  than  1  billion  tokens  ☐  1billion  to  10  trillions  tokens  ☒  More  than  10  trillions  tokens

The  text  dataset  is  a  large-scale,  multilingual  text  dataset,  comprising  highly  general  content  originating  from  publicly  available  text  datasets  and  user  data,  as  well  as  highly  specialized  and  technical  datasets,  both  synthetically  generated  and  human-curated  by  third-party  providers.
☒  Image
☐  Less  than  1  million  images  ☒  1Million  to1  billion  images  ☐  More  than  1  billion  images
The  image  dataset  is  composed  of  multimodal  content  (in  particular,  text  interleaved  with  image  data),  including  multimodal  content  sourced  from  publicly  available  datasets,  third-party  providers,  user-provided  content,  and  synthetically  generated  content.  The  data  covered  diagrams,  objects,  and  images  in  context.
☐  Audio
☐  Less  than  10  000  hours  ☐  10  000  to1  million  hours   ☐  More  than  1  million  hours
NA.
1

☐  Video
☐  Less  than  10  000  hours  ☐  10  000  to1  million  hours   ☐  More  than  1  million  hours
NA.
☐  Other
Specify  the  modality  and  for  each  one  indicate  approximate  size  and  unit  of  measurement
NA.
 Latest  date  of  data  acquisition/collection  for  model  training:
The  training  data  is  composed  of  several  datasets,  with  varying  periods  of  data  collection.  The  latest  date  of  data  collection  was  July  2025.
Description  of  the  linguistic  characteristics  of  the  overall  training  data:
 The  training  data  is  multilingual  in  its  coverage,  ensuring  broad  representation  of  official  EU  languages.   Other  relevant  characteristics  of  the  overall  training  data:
 The  overall  training  data  is  curated  and  optimized  to  improve  the  model's  multimodal,  multilingual,  and  text-generation  capabilities.   Additional  comments  (optional):  NA.

2.  List  of  data  sources
2.  List  of  data  sou r c es   2.1.  Publicly  available  datasets      Have  you  used  publicly  available  datasets  to  train  the  model?    ☒  Yes      ☐  No If  yes,  specify  the  modality(ies)  of  the  content  covered  by  the  datasets  concerned:
☒  Text    ☒  Image    ☐  Video   ☐  Audio  ☐  Other
If

so,

please

specify…
 List  of  large  publicly  available  datasets:
 The  datasets  used  to  train  the  model  include  Common  Crawl.
General  description  of  other  publicly  available  datasets  not  listed  above:
 Additionally,  Mistral  AI  used  a  combination  of  other  publicly-available  datasets  to  train  Ministral  3  -  14B.    These  datasets  covered  text  and  image  modalities,  and  included:  broad,  general-reference  datasets;  highly  specialized,  academic  datasets  or  datasets  addressing  complex  tasks  such  as  STEM,  coding  and/or  reasoning  tasks;  and  government-produced,  administrative,  or  legal  datasets.   Additional  comments  (optional):  NA.

2.2  Private  non-publicly  available  datasets  obtained  from  third  parties  2.2.1.  Datasets  commercially  licensed  by  rightsholders  or  their  representatives  Have  you  concluded  transactional  commercial  licensing  agreement(s)  with  rightsholder(s)  or  with  their  representatives?
☐  Yes      ☐ No ☒  Other
2

Mistral  AI  concludes  data  access  agreements  with  rights  holders  or  their  representatives  for  access  to  non-publicly  available  datasets.   If  yes,  specify  the  modality(ies)  of  the  content  covered  by  the  datasets  concerned:
☒  Text    ☒  Image    ☐  Video    ☐  Audio  ☐  Other
If

so,

please

specify… 2.2.2.  Private  datasets  obtained  from  other  third  parties  Have  you  obtained  private  datasets  from  third  parties  that  are  not  licensed  as  described  in  Section  2.2.1,  such  as  data  obtained  from  providers  of  private  databases,  or  data  intermediaries?
☒  Yes      ☐  No
If  yes,  specify  the  modality(ies)  of  the  content  covered  by  the  datasets  concerned:
☒  Text    ☒  Image    ☐  Video   ☐  Audio  ☐  Other  If  so,  please  specify… If  publicly  known,  list  private  datasets  obtained  from  other  third  parties:   NA.
General  description  of  non-publicly  known  private  datasets  obtained  from  third  parties
 Mistral  AI  works  with  a  variety  of  third-party  providers  for  access  to  synthetically  generated  and  human-curated  datasets.  These  datasets  are  obtained  and  curated  using  industry-standards  and  accessed  under  commercial  or  partnership  agreements  with  the  relevant  contractual  guarantees.   Additional  comments  (optional):  NA.
2.3  Data  crawled  and  scraped  from  online  sources   Were  crawlers  used  by  the  provider  or  on  behalf  of?
☒  Yes      ☐  No
If  yes,  specify  crawler  name(s)/identifier(s):  NA.
Purposes  of  the  crawler(s):
Crawlers  were  used  to  collect  publicly  available  sources  on   the  internet.
General  description  of  crawler  behaviour:
 Our  crawlers  are  designed  to  respect  robots.txt,  extract  information  from  lawfully  accessible  publicly  available  sources,  and  to  not  circumvent  technological  measures.   Period  of  data  collection:  Up  to  July  2025.
Comprehensive  description  of  the  type  of  content  and  online  sources  crawled:
 Crawlers  obtained   a  wide  range  of  publicly-accessible  content  types,  including  text,  images,  multimodal  content,  code  and  metadata   from  general-knowledge  domain  names,  and  more  specialised  websites.   Type  of  modality  covered:   ☒  Text    ☒  Image    ☐  Video   ☐  Audio
3

 ☐  Other
If

so,

please

specify…

Summary  of  the  most  relevant  domain  names  crawled:
The  most  relevant  domain  names  crawled  included  generalist  websites,  highly-specialised  resources  such  (academic  and  technical  repositories)  and  government-maintained,  administrative  or  legal  domains  (such  as  patent  portals  or  other  document-hosting  sites).   Additional  comments  (optional):  NA.
2.4  User  data   Was  data  from  user  interactions  with  the  AI  model  (e.g.  user  input  and  prompts)  used  to  train  the  model?
 ☐  Yes      ☒  No
Was  data  collected  from  user  interactions  with  the  provider’s  other  services  or  products  used  to  train  the  model?
 ☒  Yes      ☐  No
If  yes,  provide  a  general  description  of  the  provider’s  services  or  products  that  were  used  to  collect  the  user  data:
 Ministral  3  -  14B  was  trained  on  user  interaction  data,  subject  to  user  opt-out  from  training  and  strict  privacy  safeguards.  For  more  information  see  our  Privacy  Policy.   Type  of  modality  covered:  ☒  Text    ☒  Image    ☐  Video   ☐  Audio   ☐  Other
If

so,

please

specify…
 Additional  comments  (optional):  NA.
2.5  Synthetic  data   Was  synthetic  AI-generated  data  created  by  the  provider  or  on  their  behalf  to  train  the  model?
 ☒  Yes      ☐  No
If  yes,  modality  of  the  synthetic  data:
 ☒  Text    ☒  Image    ☐  Video   ☐  Audio        ☐  Other
If

so,

please

specify…
  If  yes,  specify  the  general-purpose  AI  model(s)  used  to  generate  the  synthetic  data  if  available  on  the  market:
Mistral  AI  used  its  own  models  to  generate  synthetic  data,  including  Mistral  Small  3.1  and  Mistral  Medium  3.  Information  about  other  AI  models,  including  provider’s  own  AI  model(s)  not  available  on  the  market,  used  to  generate  synthetic  data  to  train  the  model  to  which  this  Summary  applies:
 Ministral  3  -  14B  was  trained  using  synthetic  data  generated  by  other  models  such  as  internal  Mistral  AI  models.  This  data  was  combined  with  additional  synthetic  datasets  from  third-party  providers.
Additional  comments  (optional):  NA.   2.6  Other  sources  of  data  Have  data  sources  other  than  those  described  in  Sections  2.1  to  2.5  been  used  to  train  the  model?   ☐  Yes      ☒  No
4

If  yes,  provide  a  narrative  description  of  these  data  sources  and  the  data:
 NA.
Additional  comments  (optional):  NA.
1.  Data  processing  aspects  3.  Data  processing  aspects  3.1.  Respect  of  reservation  of  rights  from  text  and  data  mining  exception  or
limitation

 Are  you  a  Signatory  to  the  Code  of  Practice  for  general-purpose  AI  models  that  includes  commitments  to  respect  reservations  of  rights  from  the  TDM  exception  or  limitation?
 ☒  Yes      ☐  No
Describe  the  measures  implemented  before  model  training  to  respect  reservations  of  rights  from  the  TDM  exception  or  limitation  before  and  during  data  collection,  including  the  opt-out  protocols  and  solutions  honoured  by  the  provider  or,  as  applicable,  by  third  parties  from  which  datasets  have  been  obtained:
 Mistral  AI  enters  into  contractual  arrangements  with  third  parties  that  respect  opt-out  protocols;  Mistral  AI  uses  crawlers  that  are  designed  to  collect  from  lawfully  accessible  sources,   respect  robots.txt  instructions,  and  not  to  circumvent  technological  measures.  For  more  information,  please  see  our  website  docs.mistral.ai.   Additional  comments  (optional):
 NA.
 3.2  Removal  of  illegal  content
General  description  of  measures  taken:
Mistral  AI  implements  a  multi-layered  approach  to  avoid  or  remove  illegal  content  from  its  training  data.  At  the  data  acquisition  or  collection  stage,  Mistral  AI  sources  high-quality,  publicly  available  datasets  which  are  subject  to  industry-recognized  mechanisms  -  such  as  blacklisting,  robots.txt,  and  automated  classification  -  to  exclude  illegal  and  harmful  content.   At  the  data  processing  stage,  Mistral  AI  applies  an  additional  combination  of  model-specific  and  model-agnostic  measures,  including  proprietary  filters,  deduplication,  and  safety  classifiers  to  further  identify  and  remove  any  residual  illegal  content.
3.3.  Other  information  (optional)  Other  relevant  information  about  data  processing  (optional):
 NA.

5