GPAI Ledger › GPAI Training Transparency tracker (AI Accountability Lab (AIAL)) › Capture 12 Sep 2026

DeepSeek_DeepSeek_V4_2026_09_08 — capture 20260912T062136Z

Filed under AI Accountability Lab (AIAL) — GPAI Training Transparency tracker, the source this project captured it from; the document itself is the filing of the model named above.

Verify: sha256sum 0c630aa2165c1a4a8048c669d6e38e650ad055ca8aa5c95f7bc66546f8a96efc.pdf must equal the hash above (the filename IS the expected hash); ots verify 0c630aa2165c1a4a8048c669d6e38e650ad055ca8aa5c95f7bc66546f8a96efc.pdf.20260912T062136Z.ots -f 0c630aa2165c1a4a8048c669d6e38e650ad055ca8aa5c95f7bc66546f8a96efc.pdf (opentimestamps.org) proves the bytes existed no later than the attestation time — an upper bound on the capture time; the fetch time above is the archive's own record (a freshly captured proof reports 'pending' here: the calendars anchor within hours, but this archive only upgrades the stored proof to its anchor on a later run, so expect a day or two). ots verify needs a local Bitcoin Core node (a pruned one is fine); without one, ots info on the proof prints the attesting block height and merkle path to check on any block explorer.

Extracted text

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

1
Public Summary of Training Content for DeepSeek-V4
Version of the Summary:
Last update: 4/08/2026
General information
1. General information
1.1. Provider identification
Provider name and
contact details:
Hangzhou DeepSeek Artificial Intelligence Co., Ltd.Room 1201, Building Xi
1, Galaxy International Building, No. 169 North Huancheng Road, Gongshu
District, Hangzhou, Zhejiang, China.
Authorised representative
name and contact details:
We have appointed Prighter Group with its local partners as our authorised
representative in the European Union. Please visit:
https://app.prighter.com/portal/17162496076
1.2. Model identification
Versioned model name(s): DeepSeek-V4（including DeepSeek-V4-Pro and DeepSeek-V4-Flash. ）
Model dependencies: None
Date of placement of the
model on the Union
market: 24 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
☐ 1billion to 10 trillions tokens
☒ More than 10 trillions tokens
Publicly available internet information,
licensed datasets, and diverse textual content
including mathematical texts, source code,
multilingual materials, long-form documents
and text used for agentic and domain-
specific training.
☐ Image
☐ Less than 1 million images
☐ 1Million to1 billion images
☐ More than 1 billion images
N/A
☐ Audio
☐ Less than 10 000 hours
☐ 10 000 to1 million hours
☐ More than 1 million hours
N/A
V1
2
☐ Video
☐ Less than 10 000 hours
☐ 10 000 to1 million hours
☐ More than 1 million hours
N/A
☐ 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 DeepSeek V4 includes different datasets from
varying time periods, with some data collected approximately no later
than May 2025.
Description of the linguistic
characteristics of the overall
training data:
Other relevant characteristics
of the overall training data:
Additional comments
(optional): N/A
2. List of data sources
2. List of data sources
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:
The training data includes text from Common Crawl and Stack
Exchange. Prior to training, they are processed through quality
screening, deduplication, language identification, and safety
filtering.
General description of other
publicly available datasets not
listed above:
Other publicly available datasets include extensive multi-domain text
and code corpora released by third parties through public
repositories, online platforms, open data portals, and dedicated
websites, including encompass long documents, scientific and
technical materials, and source code. Prior to training, they are
processed through quality screening, deduplication, language
identification, and safety filtering. These datasets cover diverse
subject areas worldwide and support multiple languages.
Additional comments (optional): N/A
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
Multilingual, with strong coverage in Chinese and English.
The training corpus is designed as a whole for the text output model.
3
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: N/A
General description of non-publicly known
private datasets obtained from third parties The dataset is mainly composed of text data sourced from
multiple domains.
Additional comments (optional): N/A
2.3 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:
N/A
Type of modality covered: ☐ Text ☐ Image ☐ Video ☐ Audio
☐ Other
Additional comments (optional):
If user input is used to construct training data, we apply
secure encryption, strict de-identification, and anonymization
to make it cannot be linked to any specific individual. Users
are also given the right to opt out. For information on how to
opt out of AI training, please refer to the DeepSeek Privacy
Policy.
2.4 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
4
If yes, specify the general-purpose AI model(s)
used to generate the synthetic data if available on
the market:
DeepSeek generated synthetic data using its own AI
models, including, where applicable, publicly
available earlier DeepSeek models.
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:
DeepSeek generates synthetic data using internal
models. For instance, enriching data in areas where
specialized training data is harder to obtain,
whether by domain, language, task, or format.
Beyond that, these models can be applied to create
examples for instruction following, reasoning,
coding, multi-modal understanding, safety
alignment, and evaluation tasks.
Additional comments (optional): N/A
2.5 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: An internal team of professionals undertakes the
creation, labeling, and auditing of data across
certain domains.
Additional comments (optional): N/A
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:
DeepSeek places emphasis on compliance with
laws and regulations related to intellectual
property, trade secrets, and personal privacy. For
publicly accessible web data, DeepSeek
implements measures before and during data
collection to respect applicable rights reservations
and opt-out signals relevant to text and data
mining; the crawler is designed to respect
robots.txt instructions and other standard web
protocols, where applicable, and is not designed to
circumvent captchas, paywalls, or password-
protected content.
Additional comments (optional): N/A
3.2 Removal of illegal content
5
General description of measures taken:
DeepSeek applies preprocessing and screening measures to
screen for any proprietary information protected by intellectual
property rights or trade secrets, or any content involving
pornography, gambling, drugs, prejudice, discrimination, etc., in
order to avoid or remove illegal content from the training data.
These measures may include automatic filtering, manual
inspection, keyword inspection, and classification model
inspection methods.
3.3. Other information (optional)
Other relevant information about data
processing (optional):
N/A