GPAI Ledger › GPT Image 2 (OpenAI) › Capture 11 Aug 2026

GPT Image 2 — capture 20260811T111046Z

Verify: sha256sum 8a3ac329e248433d3a2db555236fcc94684c0294d2a705bbaf2a60995c216b3a.pdf must equal the hash above (the filename IS the expected hash); ots verify 8a3ac329e248433d3a2db555236fcc94684c0294d2a705bbaf2a60995c216b3a.pdf.ots -f 8a3ac329e248433d3a2db555236fcc94684c0294d2a705bbaf2a60995c216b3a.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
ChatGPT

Images

2.0
  Version  of  the  Summary:   v1  Last  update:   30  July  2026  General  information
1.  General  information  1.1.  Provider  identification
Provider  name  and  contact  details:
OpenAI  Ireland  Ltd,  1st  Floor,  The  Liffey  Trust  Centre,  117-126  Sheriff  Street  Upper,  Dublin  1,  D01  YC43,  Ireland.
Authorised  representative  name  and  contact  details:
Not  applicable  (provider  established  in  the  Union)
 1.2.  Model  identification   Versioned  model  name(s):                          ChatGPT  Images  2.0
Model  dependencies:
ChatGPT  Images  2.0  is  not  a  modification  or  a  fine-tune  of  a  prior  model.  The  disclosures  below  regarding  types  and  quantities  of  training  data  apply  to  ChatGPT  Images  2.0  and  all  subsequent  releases  in  the  model  lifecycle.  Date  of  placement  of  the  model  on  the  Union  market:  21  April  2026
1.3  Modalities,  overall  training  data  size  and  other  characteristics
Modality  Select  the  modalities  present  in  the  training  data,  to  the  extent  that
they

are

identifiable

Training  data  size  For  each  selected  modality,  select  the  range  within  which  the  estimated  total  training  data  size  for  that  modality  falls.  Dynamic  datasets  may  be  excluded  from  the  estimation.
Types  of  content  For  each  selected  modality,  provide  a  general  description  of  the  type  of  content
that

has

been

included

in

the

training

data.

☒  Text
☐  Less  than  1  billion  tokens  ☐  1  billion  to  10  trillion  tokens  ☒  More  than  10  trillion  tokens

ChatGPT  Images  2.0  was  trained  on  a  large-scale,  multilingual  mixture  of  publicly  available  data,  data  accessed  through  partnerships,  synthetic  data,  and  human-generated  text,  including  general  web  content,  reference  materials,  technical  documentation,  source  code,  and  other  text,  curated  and  filtered  for  quality  and  safety.
☒  Image
☐  Less  than  1  million  images  ☐  1  million  to  1  billion  images  ☒  More  than  1  billion  images
ChatGPT  Images  2.0  was  trained  on  a  broad  corpus  of  images  from  publicly  available  sources,  images  accessed  through  partnerships,  and
1

synthetic  images,  spanning  natural  scenes,  objects,  people,  and  diagrams,  with  filtering  to  support  quality,  safety,  and  multimodal  understanding.
☒  Audio
☐  Less  than  10  000  hours  ☐  10  000  to  1  million  hours  ☒  More  than  1  million  hours
ChatGPT  Images  2.0  was  trained  on  a  diverse  set  of  publicly  available  audio  data,  including  conversational,  narrated,  instructional,  and  informational  material,  represented  in  part  through  transcripts,  and  curated  and  filtered  for  quality  and  safety.
☒  Video
☐  Less  than  10  000  hours  ☐  10  000  to  1  million  hours  ☒  More  than  1  million  hours
ChatGPT  Images  2.0  was  trained  on  publicly  available  multi-domain  audiovisual  content,  including  spoken,  narrated,  instructional,  and  general-interest  audiovisual  material,  represented  in  part  through  transcripts,  captions,  or  metadata,  and  filtered  for  quality  and  safety.
☐  Other
Specify  the  modality  and  for  each  one  indicate  approximate  size  and  unit  of  measurement
N/A
 Latest  date  of  data  acquisition/collection  for  model  training:
The  data  used  to  train  ChatGPT  Images  2.0  includes  different  datasets  from  varying  time  periods,  with  some  data  collected  no  later  than  April  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:
ChatGPT  Images  2.0  is  an  image-generation  model  trained  on  a  broad  multimodal  corpus.  Its  training  data  includes  text,  images,  video-derived  visual  material,  documents  and  diagrams,  paired  captions  and  metadata,  and  synthetic  visual  and  text  data.  This  mix  supports  understanding  prompts  and  visual  context  and  generating  and  editing  images  across  a  wide  range  of  subjects,  styles,  languages,  and  formats.  Additional  comments  (optional):  N/A
2.  List  of  data  sources
2.  List  of  data  sources   2.1.  Publicly  available  datasets
2

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:   The  training  data  for  ChatGPT  Images  2.0  includes  text  from  Common  Crawl.
General  description  of  other  publicly  available  datasets  not  listed  above:
Other  publicly  available  datasets  include  broad,  multi-domain  text  and  image  datasets  made  available  by  third  parties  through  public  repositories,  online  platforms,  and  specialized  websites,  including  reference  materials,  scientific  and  technical  content,  source  code,  image-text  datasets,  and  audiovisual  datasets  distributed  with  captions,  transcripts,  metadata,  or  related  text.  These  datasets  are  global  in  scope,  multilingual,  and  subject  to  preprocessing  such  as  quality  filtering,  deduplication,  and  safety  filtering  before  training.  We  use  advanced  data  filtering  processes  to  reduce  personal  information  from  training  data.
Additional  comments  (optional):
From  public  web  datasets,  OpenAI  takes  steps  to  identify  and  apply  relevant  rights-reservation  and  opt-out  signals,  including  robots.txt  signals  for  GPTBot,  where  those  signals  are  available  for  domains  listed  in  the  datasets.  OpenAI  also  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.
2.2  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  (see  below)
If  yes,  specify  the  modality(ies)  of  the  content  covered  by  the  datasets  concerned:
☒  Text    ☒  Image    ☐  Video    ☐  Audio    ☐  Other
Additional  comments  (optional):
OpenAI  enters  into  broad  partnerships  with  third  parties  that  may  include,  among  other  initiatives,  rights  to  display  partner  content  to  users  in  our  products  and/or  access  to  non-publicly  available  content,  such  as  archives  and  metadata.  OpenAI  does  not  pursue  partnerships  solely  for  access  to  publicly  available  data.
3

 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:
N/A
If  publicly  known,  list  private  datasets  obtained  from  other  third  parties:   N/A  General  description  of  non-publicly  known  private  datasets  obtained  from  third  parties
N/A
Additional  comments  (optional):  N/A
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):  GPTBot
Purposes  of  the  crawler(s):
GPTBot  is  used  to  crawl  content  that  may  be  used  to  train  OpenAI’s  generative  AI  foundation  models  and  make  them  more  useful  and  safe.  OpenAI  publishes  additional  information  about  its  crawlers,  their  behavior,  user-agent  identifiers,  and  IP  addresses  at  https://developers.openai.com/api/docs/bots.
General  description  of  crawler  behaviour:
OpenAI’s  crawler  is  designed  to  respect  robots.txt  instructions  for  the  GPTBot  user-agent,  including  instructions  indicating  that  crawled  content  should  not  be  used  to  train  OpenAI’s  generative  AI  foundation  models.  OpenAI’s  crawler  is  not  designed  to  circumvent  captchas  or  paywalls  or  to  access  password-protected  content.  OpenAI  also  filtered  the  training  data  for  ChatGPT  Images  2.0  for  domains  that  have  been  recognized  as  persistently  and  repeatedly  infringing  copyright,  using  the  U.S.  Trade  Representative  (USTR)  Notorious  Markets  for  Counterfeiting  and  Piracy  list  as  a  signal.  Period  of  data  collection:  Approximately  2018  –  January  2026
Comprehensive  description  of  the  type  of  content  and  online  sources  crawled:
Crawled  content  includes  a  broad  range  of  publicly  accessible  online  material,  including  reference,  educational,  scientific,  technical,  government  and  institutional,  and  general-interest  content.  Crawled
4

sources  include  text,  images,  and  associated  metadata  such  as  captions,  alt  text,  transcripts,  or  other  descriptive  text,  and  were  filtered  for  quality  and  safety  before  training.  Type  of  modality  covered:  ☒  Text    ☒  Image    ☒  Video    ☒  Audio    ☐  Other   Summary  of  the  most  relevant  domain  names  crawled:
The  most  relevant  crawled  source  domains  include  broad  multilingual  webpages  and  image-rich  sites,  document  and  PDF  repositories,  and  audiovisual  media.  Additional  comments  (optional):  N/A
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:
Subject  to  privacy  settings,  controls,  user  requests  and  opt-outs,  and  our  policies,  for  individuals  using  products  such  as  ChatGPT,  and  Codex,  OpenAI  may  use  interactions  to  train  our  models,  including  Images  2.0.  More  information  on  OpenAI's  policies  is  located  here:  https://help.openai.com/en/articles/5722486-how-your-d  ata-is-used-to-improve-model-performance.   Additionally,  OpenAI's  Privacy  Portal,  which  contains  information  regarding  OpenAI's  privacy  policies,  is  located  here:  https://privacy.openai.com/policies.    Type  of  modality  covered:  ☒  Text    ☒  Image    ☐  Video    ☐  Audio    ☐  Other
Additional  comments  (optional):
We  use  advanced  data  filtering  processes  to  reduce  personal  information  from  training  data.   Relevant  public  information  on  user  controls  is  available  here:  https://help.openai.com/en/articles/5722486-how-your-d  ata-is-used-to-improve-model-performance
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,  and  audio  If  yes,  specify  the  general-purpose  AI  model(s)  used  to  generate  the  synthetic  data  if  available  on  the  market:
N/A.
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:
OpenAI  uses  specialized  internal  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,  coding,  multimodal  understanding,  image  generation,  image  annotation,  and  improving  safety.   Additional  comments  (optional):  N/A    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:
OpenAI  and  our  vendors  create  data  to  help  our  models  improve  on  a  wide  variety  of  tasks.  For  example,  OpenAI  worked  with  human  trainers  to  create  and  annotate  data  for  image  generation  and  editing.  This  included  comparative  assessments  of  model  outputs,  which  were  used  during  post-training  to  improve  image  quality  and  how  well  the  model  follows  generation  and  editing  instructions..   Additional  comments  (optional):  N/A
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:
OpenAI  implements  measures  to  respect  rights  reservations  and  opt-out  signals  relevant  to  text  and  data  mining.  For  web  data  used  for  training,  OpenAI’s  crawler  is  designed  to  respect  robots.txt  instructions  for  the  GPTBot  user-agent,  including  instructions  indicating  that  crawled  content  should  not  be  used  to  train  OpenAI’s  generative  foundation  models.  OpenAI’s  crawler  is  not  designed  to  circumvent  captchas  or  paywalls  or  to  access  password-protected  content.  Additional  comments  (optional):  N/A
6

 3.2  Removal  of  illegal  content
General  description  of  measures  taken:
OpenAI  applies  preprocessing  and  screening  measures  intended  to  avoid  or  remove  illegal  content  under  Union  law  from  training  data.  These  may  include  automated  filtering,  keyword-based  rules,  hash-matching,  and  model-based  classifiers  to  help  identify  and  exclude  unlawful  material.
3.3.  Other  information  (optional)
Other  relevant  information  about  data  processing  (optional):
N/A
7