GPAI Ledger › Grok Voice Think Fast 2.0 (xAI) › Capture 18 Aug 2026

Grok Voice Think Fast 2.0 — capture 20260818T164338Z

Verify: sha256sum 03b534c895b34039ecd18a7d94495626b24ca3286797da617b0666acb4e33d97.pdf must equal the hash above (the filename IS the expected hash); ots verify 03b534c895b34039ecd18a7d94495626b24ca3286797da617b0666acb4e33d97.pdf.ots -f 03b534c895b34039ecd18a7d94495626b24ca3286797da617b0666acb4e33d97.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  for  Voice   Version  of  the  Summary:   V1  Last  update:   29  July  2026  General  information
1.  General  information  1.1.  Provider  identification   Provider  name  and  contact  details:
xAI  LLC,  1450  Page  Mill  Rd,  Palo  Alto,  CA  94304
Authorised  representative  name  and  contact  details:
EDSR,  Valukoja  8/2,  2nd  floor,  Tallinn,  Estonia,  EE-11415
1.2.  Model  identification   Versioned  model  name(s):  Grok  Voice  Think  Fast  2.0  Model  dependencies:  N/A Date  of  placement  of  the  model  on  the  Union  market:  29  July  2026

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

Grok  Voice  Think  Fast  2.0  was  trained  using  a  carefully  designated  data  recipe  that  incorporates  a  diverse  corpus  of  publicly  available  information,  including,  for  example,  certain  scientific  text,  legal  and  official  documents,  and  X  social  media  posts.  The  training  data  also  encompasses  content  produced  by  third  parties,  including  data  from  users  and  contractors,  as  well  as  internally  generated  data.  Training  data  was  curated  and  filtered  for  quality  and  safety.
☐  Image
☐  Less  than  1  million  images  ☐  1Million  to1  billion  images  ☐  More  than  1  billion  images
N/A
☒  Audio
 1

☐  Less  than  10  000  hours  ☐  10  000  to1  million  hours   ☒  More  than  1  million  hours
Grok  Voice  Think  Fast  2.0  was  trained  using  a  carefully  designated  data  recipe  that  incorporates  a  diverse  corpus  of  publicly  available  audio  data,  including,  for  example,  public  speech,  media,  voice  application  recordings,  noise,  audio  communications,  certain  clips  and  audio  from  social  media  videos.  The  training  data  may  also  encompass  audio  content  produced  by  third  parties,  including  from  users  and  contractors,  as
1
 Excluding  audio  that  is  part  of  video,  as  this  should  be  reported  under  the  “video”  modality  instead.  Furthermore,  the  Commission
understands

the

modality

of

‘audio’

to

include

‘speech’.

1

well  as  internally  generated  content.  Training  data  was  curated  and  filtered  for  quality  and  safety.
☒  Video
☒  Less  than  10  000  hours  ☐  10  000  to1  million  hours   ☐  More  than  1  million  hours
Grok  Voice  Think  Fast  2.0  was  trained  on  publicly  available  audiovisual  content,  including,  for  example,  certain  clips  and  X  social  media  videos.   The  training  data  may  also  encompass  audiovisual  content  produced  by  third  parties,  including  from  users  and  contractors,  as  well  as  internally  generated  content.  Training  data  was  curated  and  filtered  for  quality  and  safety.
☐  Other
N/A  N/A
 Latest  date  of  data  acquisition/collection  for  model  training:
The  data  used  to  train  Grok  Voice  Think  Fast  2.0  includes  different  datasets  from  varying  time  periods,  with  data  collected  no  later  than  June  2026.   Description  of  the  linguistic  characteristics  of  the  overall  training  data:
Multilingual,  with  strong  English  coverage  and  substantial  representation  across  EU  official  languages  and  other  languages  from  around  the  world.

Other  relevant  characteristics  of  the  overall  training  data:
Grok’s  training  corpus  is  designed  for  a  highly  speech-to-speech  model  with  strong  multimodal  understanding.  It  draws  from  a  range  of  written  material,  audio  and  audiovisual  content.  The  corpus  is  intentionally  broad,  aiming  for  extensive  coverage  across  topics,  geographies,  languages,  and  formats.
Additional  comments  (optional):

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
●  List  of  large  publicly  available  datasets:
●  Common  Voice:  a  free,  open-source  crowdsourcing  project  started  by  Mozilla  to  create  a  public  domain  database  of  diverse  human  voices.  It  helps  developers  train  speech-to-text  and  text-to-speech  AI  applications  without  high  costs  or  proprietary  (dates  of  data  collection  not  known)  ●  LibriSpeech:  LibriSpeech  is  a  large  speech  dataset  containing  about  1,000  hours  of  English  audio,  designed  for  training  and  testing  speech  recognition  systems.  The  audio  comes  from  read  audiobooks  in  the LibriVox project  and  is  sampled  at  16  kHz  (dates  of  data  collection  not  known)
2

●  VoxPopuli:  VoxPopuli  is  a  large-scale  multilingual  speech  corpus  for  representation  learning,  semi-supervised  learning  and  interpretation.  The  raw  data  is  collected  from  2009-2020  European  Parliament  event  recordings.  ●  Fleurs :  speech  version  of  the  FLoRes  machine  translation  benchmark.  Uses  2009  n-way  parallel  sentences  from  the  FLoRes  dev  and  devtest  publicly  available  sets,  in  102  languages  (dates  of  data  collection  not  known).
General  description  of  other  publicly  available  datasets  not  listed  above:
 Other  publicly  available  datasets  used  by  Grok  Voice  Think  Fast  2.0  consisted  of  text  and  audio,  content  and  audiovisual  content,  in  various  different  languages,  from  other  publicly  available  sources  (dates  of  data  collection  not  known).   Additional  comments  (optional):
Audio  data  is  packaged  for  diverse  audio  processing  tasks  (speech  recognition,  speaker  recognition,  speech  generation,  acoustic  event  analysis)
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
If  yes,  specify  the  modality(ies)  of  the  content  covered  by  the  datasets  concerned:
☐  Text    ☐  Image    ☐  Video    ☐  Audio  ☐  Other
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  publicly  known,  list  private  datasets  obtained  from  other  third  parties:
General  description  of  non-publicly  known  private  datasets  obtained  from  third  parties
  Private  datasets  used  to  train  Grok  Voice  Think  Fast  2.0  that  are  not  publicly  known  consisted  of  audio,  including  conversations  on  various  different  topics,  recorded  in  clean  and  high  quality  environments.  Additional  comments  (optional):
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):  xAI  Web  Crawler
3

Purposes  of  the  crawler(s):
 xAI  uses  a  web  crawler  to  crawl  certain  publicly  available  internet  pages  to  help  improve  language  understanding,  reasoning,  and  overall  capabilities.  General  description  of  crawler  behaviour:
 xAI  uses  a  web  crawler  to  discover  and  scan  certain  publicly  available  internet  pages  and  rescan  for  updates  and  analysis. Period  of  data  collection:  From  January  2024  to  June  2026
Comprehensive  description  of  the  type  of  content  and  online  sources  crawled:
Content  encompasses  a  wide  variety  of  publicly  available  internet  data  including  educational,  scientific,  technical,  government,  and  institutional  sources.  These  materials  may   consist  of  text,  audio,  and  video.  Type  of  modality  covered:   ☒  Text    ☐  Image    ☒  Video   ☒  Audio   ☐  Other
Summary  of  the  most  relevant  domain  names  crawled:
The  domains  accounting  for  the  largest  share  of  content  crawled  and  used  to  train  Grok  Voice  Think  Fast  2.0,  may  comprise  a  mix  of  publicly  available  information,  including  larger  social  media  and  content  sharing  platforms,  educational  repositories,  government  and  institutional  resources,  community  and  document-sharing  sites.  Audio  and  audiovisual  content  may  be  accompanied  by  publicly  available  transcripts,  captions,  titles,  descriptions,  and  other  associated  metadata.  This  associated  text  provides  contextual  and  linguistic  information  relevant  to  speech  recognition,  speech  generation,  and  multilingual  understanding.
Additional  comments  (optional):
xAI  uses  the  U.S.  Trade  Representative  (USTR)  Notorious  Markets  for  Counterfeiting  and  Piracy  list  as  a  signal  when  deciding  to  exclude  data  from  certain  websites  that  have  been  recognized  as  persistently  and  repeatedly  infringing  copyright.
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:

Type  of  modality  covered:  ☒  Text    ☐  Image    ☐  Video   ☒  Audio  ☐  Other
Additional  comments  (optional):
Subject  to  privacy  settings,  controls,  user  requests,  opt-outs,  and  our  policies,  xAI  may  use  interactions  with  Grok  to  help  train  and  improve  our  models,  including  Grok  Voice  Think  Fast  2.0.  For  more  information  on  how  we  use  data  to  improve  model  performance,  please  visit: https://x.ai/privacy.  xAI  uses  advanced  data  filtering  processes  to  reduce  personal  information  from  training  data.  We  also  use  advanced  data  processes  to  reduce  the  amount  of  personal  data  in  our  training  data.
4

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  yes,  specify  the  general-purpose  AI  model(s)  used  to  generate  the  synthetic  data  if  available  on  the  market:
xAI  generates  synthetic  data  using  its  own  general-purpose  AI  models,  including  Grok  4.3,  Grok  4.5.
 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:
 xAI  uses  other  models  to  generate  synthetic  data  for  targeted  training  objectives,  including  augmenting  data  in  domains,  languages,  tasks,  or  formats  where  specialized  training  data  is  comparatively  scarce.  These  models  may  be  used  to  generate  examples  for  instruction  following,  reasoning,  coding,  multimodal  understanding,  safety,  and  evaluation.
Additional  comments  (optional):
xAI’s  tutors  used  xAI’s  own  general-purpose  AI  models  to  synthetic  data,  text  and  audio  conversations  that  could  be  used  to  train  Grok  Voice  Think  Fast  2.0.    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
If  yes,  provide  a  narrative  description  of  these  data  sources  and  the  data:
 xAI  also  creates  training  data  through  employees  and  tutors  where  suitable  existing  data  is  not  available.  For  Grok  Voice  Think  Fast  2.0,  this  included  human  recorded  conversations  and  simulated  interactions  across  a  range  of  topics  and  contexts.   Additional  comments  (optional):
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
5

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:
 xAI  implements  measures  to  respect  rights  reservations  and  opt-out  signals  relevant  to  data.  For  example,  xAI  does  not  train  on  X  data  from  X  users  who  have  opted  out  of  training.  For  web  data  used  for  training,  xAI  respects  instructions  that  identified  crawled  content  should  not  be  used  to  train  xAI’s  generative  foundation  models.   Additional  comments  (optional):
 3.2  Removal  of  illegal  content
General  description  of  measures  taken:
xAI  applies  preprocessing  and  screening  measures  designed  to  avoid  or  remove  content  that  may  be  illegal  from  our  training  data.  These  measures  include  filtering,  keyword-based  rules  and  model-based  classifiers.   3.3.  Other  information  (optional)  Other  relevant  information  about  data  processing  (optional):

6