GPAI Ledger › Apertus v1.5 (Swiss AI Initiative) › Capture 17 Aug 2026

Apertus v1.5 — capture 20260817T080527Z

Note: this is a GPAI Code of Practice document (Art. 53(1)(b)–(c) — model documentation and copyright policy), not the Art. 53(1)(d) public training-data summary.

Verify: sha256sum 2e4b304b04e445632f074ae10481386a72e97945c8a6b8ae8788a73f2b3c72e5.pdf must equal the hash above (the filename IS the expected hash); ots verify 2e4b304b04e445632f074ae10481386a72e97945c8a6b8ae8788a73f2b3c72e5.pdf.ots -f 2e4b304b04e445632f074ae10481386a72e97945c8a6b8ae8788a73f2b3c72e5.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).

Code  of  Practice  for  the  Apertus  LLM  (GPAI)
This  document  contains  the  Code  of  Practice  for  the  Apertus  LLM  published  by  the  Swiss
National

AI

Institute

(SNAI),

a

partnership

between

the

two

Swiss

Federal

Institutes

of

Technology,

ETH

Zurich

and

EPFL,

on

17

July

2026.

The  model  is  released  under  a  free  and  open-source  license  that  allows  for  the  access,
usage,

modification,

and

distribution

of

the

model,

and

whose

parameters,

including

the

weights,

the

information

on

the

model

architecture,

and

the

information

on

model

usage,

are

made

publicly

available.

Despite

the

exceptions

for

GPAI

models

released

under

these

conditions,

SNAI

provides

this

Code

of

Practice

voluntarily

for

transparency

purposes

and

for

the

users’

convenience
.

The

Swiss

Federal

Institutes

have

not

signed

the

signature

form

provided

by

the

EU

AI

Office.

The  Apertus  LLM  is  a  general-purpose  AI  model  without  foreseeable  systemic  risk.
Therefore,

this

Code

of

Practice

comprises

two

(2)

chapters:

A)

Transparency

and

B)

Copyright

Policy.

Release  Version  of  this  CoP:  v1.5
Last

update:

17.07.2026

A)  TRANSPARENCY  DOCUMENTATION  (Art.  53(1)(b)  EU  AI  Act
The  following  information  outlines  the  measures  regarding  transparency.

1)  General  Information
Legal  name  for  the  model  provider:  Both  Swiss  Federal  Institutes  of  Technology,  ETH
Zurich

and

EPFL,

are

cooperation

partners

of

the

Swiss

National

AI

Institute

(SNAI)

https://www.swiss-ai.org/ llm-requests@swiss-ai.org
Authorised  representative  name  and  contact  details  (in  EU):  N/A
Model

family:

Apertus

Versioned  model  name:  Apertus  1.5
Release

date:

14.

July

2026

Union  market  release:  20.  July  2026
Model  dependencies:  Apertus  v1.

2)  Model  Properties
Model  architecture:  Decoder-only  transformer  architecture  with  xIELU  activations,  QK-
Norms,

Pre-Norm

by

RMSNorms,

and

rotary

positional

embeddings.

Input  modalities:  Text,  audio,  and  image
Maximum  input  size:  Native  context  length  64k  tokens,  extensible  to  128k
Output

modalities:

Text-only

Maximum  output  size:  N/A  (DP  dependent)
Total  model  size:  70  billion  parameters,  and  8  billion  respectively

3)  Methods  of  Distribution  and  Licenses
Distribution  channels:  Hugging  Face
Model  License:  Apache  2.0  (January  2004),  accessible  at  https://www.apache.org/licenses/LICENSE-2.0
Additional  assets  made  available,  incl.  description  of  access  and  additional  licenses:
-  Data  processing  code:
o  github.com/swiss-ai/pretrain-data
o  github.com/swiss-ai/posttrain-data
o  github.com/swiss-ai/multimodal-data
-  Model  pretraining  code:
o  github.com/swiss-ai/pretrain-code
-  Base  &  instruct  models  as  well  as  intermediate  training  checkpoints
o  huggingface.co/swiss-ai/Apertus-v1.5-8B
o  huggingface.co/swiss-ai/Apertus-v1.5-70B

4)  Use
Acceptable  Use  Policy:  See  huggingface.co/swiss-ai/Apertus-v1.5-70B  Intended  uses:  General-purpose  AI  model
Type  and  nature  of  AI  systems  in  which  the  general-purpose  AI  model  can  be  integrated:
Conversational

AI

Systems,

AI

Workflows,

Research

&

Development

Tools

Technical  means  for  model  integration:  Support  for  common  inference  frameworks,  such
as

vLLM,

SGlang,

Hugging

Face

Transformers.

See

model

card

on

Hugging

Face

for

full

details.

Required  hardware:  N/A
Required

software:

N/A

Export

Regulations:

N/A

5)  Information  on  Data  Used  for  Training,  Testing,  and  Validation
Training  Data  Type/Modality:  Text,  audio,  and  image
Latest  date  of  data  acquisition:  Main  pretraining  dataset  knowledge  cutoff  is  03/2024,
while

some

domain-specific

parts

of

the

dataset

(math)

and

parts

of

the

post-training

datasets

have

a

later

date

of

collection.

Training  Data  Provenance:  The  following  large  pretraining  datasets  derived  from
CommonCrawl

were

used,

with

data

from

2013

onward

to

a

knowledge

cut-off

mainly

of

March

2024

(Apertus

v1),

though

with

smaller

newer

documents

added

from

2025

and

2026

for

Apertus

v1.5

(the

newest

being

from

28.

April

2026).

Datasets

were

not

used

in

raw

form

but

additionally

filtered

for

opt-out

retrospectively,

for

toxicity,

high

quality,

and

other

preprocessing

as

detailed

below.

We  refer  to  the  Apertus  v1  technical  report  for  the  list  of  training  datasets  for  v1.
Apertus

v1.5

is

a

continuous

pretraining

of

v1,

on

the

following

additional

large
 datasets.  The  complete  list  of  training  datasets  is  accessible  on  our  GitHub
Text  Datasets
HuggingFaceFW/fineweb-2:  Large-scale,  high-quality  multilingual  web  text  corpus  derived
from

filtered

Common

Crawl

data.

Serves

as

a

primary

general-purpose

pretraining

source

for

LLMs,

with

strong

emphasis

on

diversity

and

reduced

noise.

License:

ODC-BY.

(Subject

to

PII

removal

and

robots.txt

filtering.)

HuggingFaceTB/dclm-edu:  Curated  educational  web  dataset  from  the  DataComp-LM
project.

Provides

high-signal,

knowledge-rich

text

optimised

for

reasoning

and

factual

learning

in

language

models.

License:

CC-BY-4.0.

(PII

removal

and

robots.txt

filtering

applied.)
nvidia/Nemotron-CC-v2.1:  NVIDIA-curated  Common  Crawl  dataset  (v2.1)  featuring
quality

scoring

and

filtering

for

large-scale

LLM

pre-training.

License:

Nvidia’s

custom

data

and

model

license

(permissive,

see

dataset

page

for

terms).

HuggingFaceFW/finePDFs-edu:  High-quality  collection  of  educational,  scientific,  and
technical

PDF

documents

with

extracted

clean

text.

Enhances

long-context

and

domain

knowledge

capabilities.

License:

ODC-BY.

(Filtered

for

quality

and

compliance.)

joelniklaus/Multi_Legal_Pile:  Specialized  multilingual  legal  corpus  aggregating  statutes,
case

law,

contracts,

and

regulatory

texts

from

multiple

jurisdictions.

Strengthens

legal

reasoning

and

domain

adaptation.

License:

only

compliant

(non-SA,

non-NC)

subsets

of

this

compound

dataset

were

used.

(PII

removal

and

robots.txt

filtering

applied.)

nvidia/Nemotron-Pretraining-Code-v1:  Large-scale,  curated  code  corpus  from  NVIDIA
designed

to

boost

programming,

software

engineering,

and

logical

reasoning

abilities.

License:

Nvidia’s

custom

data

and

model

license

(permissive,

see

dataset

page

for

terms).

Audio  Datasets
mozilla/CommonVoice24:  Mozilla’s  crowdsourced  multilingual  speech  corpus  with
validated

transcriptions

across

many

languages

and

accents.

A

cornerstone

dataset

for

inclusive,

robust

automatic

speech

recognition

(ASR)

and

text-to-speech

(TTS).

License:

CC-BY-1.0.

speechcolab/gigaspeech:  Large-scale  English  speech  recognition  corpus  (~10k  hours)
sourced

from

audiobooks,

podcasts,

and

YouTube

with

high-quality

transcriptions.

Supports

general-domain

ASR

and

audio

understanding.

License:

Apache-2.0.

MLCommons/peoples_speech:  One  of  the  largest  publicly  available  multilingual  speech
datasets,

containing

tens

of

thousands

of

hours

of

diverse,

real-world

speech.

License:

CC-BY-SA

/

CC-BY.

facebookresearch/voxpopuli:  Large  multilingual  speech  corpus  extracted  from  European
Parliament

sessions,

covering

numerous

EU

languages

with

aligned

transcripts.

Excellent

for

cross-lingual

and

parliamentary-domain

speech

tasks.

License:

CC-BY-1.0.

k2-fsa/libriheavy:  Massive  clean  English  read-speech  corpus  (tens  of  thousands  of  hours)
built

on

LibriSpeech

audiobooks

with

precise

alignments.

Known

for

high

acoustic

quality

and

utility

in

ASR/TTS

research.

License:

Apache-2.0.

facebook/omnilingual-asr-corpus:  Massive  multilingual  speech  corpus  spanning  a  wide
range

of

languages

and

dialects,

created

to

advance

open

ASR

systems

globally.

License:

CC

BY

4.0.

Image  Datasets
mlfoundations/MINT-1T:  Landmark  large-scale  image-text  dataset  scaled  to  trillions  of
tokens,

specifically

designed

to

dramatically

expand

open-source

multimodal

pretraining

data.

License:

CC-BY-4.0.

(Includes

PII

removal

and

robots.txt

filtering.)

UCSC-VLAA/Recap-DataComp-1B:  Billion-scale  image-text  dataset  featuring  high-quality

recaptions  of  the  original  DataComp-1B  collection.  Optimized  for  vision-language
pretraining

and

detailed

visual

understanding.

License:

CC-BY-4.0.

dclure/laion-aesthetics-12m-umap:  Curated  12-million-image  subset  of  LAION  focused  on
high

aesthetic

quality

(via

CLIP

and

aesthetic

scoring).

Popular

for

training

visually

pleasing

image

understanding

and

generation

models.

License:

MIT.

mvp-lab/LLaVA-OneVision-1.5-Mid-Training-85M:  Large-scale  (85M  samples)  mid-training
dataset

for

vision-language

models,

emphasizing

instruction

tuning

and

multimodal

alignment.

License:

Apache-2.0.

DeepGlint-AI/DanQing100M:  Large-scale  Chinese  image-text  pretraining  dataset  (100M
pairs)

supporting

enhanced

multilingual

and

culturally

relevant

visual-language

capabilities.

License:

CC-BY-4.0.

UCSC-VLAA/MedTrinity-25M:  Large  medical  image-text  dataset  (25M  samples)  with  rich
annotations

and

captions.

Key

resource

for

building

specialized

medical

vision-language

understanding

and

diagnostic

assistance

features.

License:

only

compliant

(non-SA,

non-NC)

subsets

of

this

compound

dataset

were

used

(see

full

list

on

the

dataset’s

page).

Other  smaller  publicly  available  and  permissively  licensed  datasets  were  used  for  the
purposes

described

above.

The

exhaustive

list

of

pre-training

datasets

is

accessible

on
 our  GitHub.  Generally,  all  republished  datasets  used  for  Apertus  v1.5  will  be  made  publicly  available  in  the  dedicated  collection  on  Hugging  Face.

Data  curation  methodologies:  See  Technical  Report.

B)  COPYRIGHT  POLICY  (Art.  53(1)(c)  EU  AI  Act)
The  following  information  outlines  the  measures  of  taken  to  comply  with  copyright.  In
particular

it

identifies,

and

complies

with,

reservation

of

rights

expressed

by

rightsholders.

1)  Reproduction  and  extraction  of  lawfully  accessible  copyright-protected
content

SNAI  has  implemented  and  adheres  to  the  following  measures  to  reduce  and  extract
only

lawfully

accessible

content

for

the

training

Apertus

LLM:

−
 The  pre-training  data  used  for  the  Apertus  LLM  was  obtained  in/from  datasets
licensed

from

Hugging

Face,
 Open  SLR,  AISHELL-1,  FHNW  Institute  for  Data
Science

Datasets,

Common

Voice,

Zenodo

(CERN),

NDL

Lab,

Swisstopo,

Geoservices,  Geodaten  BGDI,  Kaggle,  Figshare,  HoloAssist,  Our  World  in  Data,
NASA,

Smithonians,

National

Library

of

Medicine (for  details  please  refer  to  chapter
A,

above,

and

specifically

the

dataset

list

in

the

Public

Summary).

SNAI

respects

technological

denial

and/or

restriction

of

access

imposed

by

copyright

holders.

No

content

from

subscription

models

or

paywalls

was

used

(or

circumvented

for

access

purposes)

for

training

of

the

Apertus

LLM.

To  the  best  of  knowledge,  the  Apertus  LLM  was  not  trained  on  data  recognised  as
persistently

and

repeatedly

infringing

copyright

and

related

rights

on

a

commercial

scale

by

courts

or

public

authorities

in

the

European

Union

and

the

European

Economic

Area.

2)  Identification  and  compliance  with  rights  reservations
SNAI  has  identified  and  complied  with  rights  reservations,  including  through  state-of-
the-art

technologies

and

machine-readable

reservations,

e.g.

robots.txt.

The

Apertus

v1.5

LLM

was

trained

on

web

documents

crawled

by

CommonCrawl

while

respecting

standard

machine-readable

opt-out

by

websites.

In

addition,

data

from

websites

which

have

recently

opted

out

by

specifying

at

least

one

of

the

common

AI

crawlers,

at

the

time

of

January

2025,

was

removed.

Crucially,

such

removals

were

also

applied

retroactively

in

all

earlier

crawls

since

2013,

of

each

corresponding

website

present

in

our

datasets.

Pretraining

and

posttraining

datasets

were

additionally

filtered

for

licence

compliance,

and

processed

by

PII

removal

(for

details

please

refer

to

chapter

A,

above).

3)  Mitigation  of  the  risk  of  copyright-infringing  outputs
To  mitigate  the  risk  that  a  downstream  AI  system,  into  which  the  Apertus  LLM  (AI  model)
is

integrated,

generates

output

that

may

infringe

copyrights,

SNAI

−  implemented  state-of-the-art  mitigation  techniques  to  avoid  verbatim
memorization

in

the

model,

by

the

Goldfish

loss

technique

https://arxiv.org/html/2406.10209

,

which

avoids

verbatim

memorization

of

text

sequences

longer

than

50

tokens.

More

precisely,

every

50th

token

(on

average)

of

the

pretraining

data

is

not

provided

a

prediction

target,

i.e.

has

no

loss

function,

and

thus

breaks

any

verbatim

memorization

beyond

that

sequence

length.

More

detailed

results

on

the

success

of

this

mitigation

technique

is

provided

in

the

model’s

technical

report.

SNAI

considers

these

measures

as

appropriate

and

proportionate

technical

safeguards

to

prevent

the

Apertus

LLM

from

generating

outputs

that

reproduce

training

content

(for

details

please

refer

to

chapter

A,

above);

provided,

however,

that

downstream

providers

remain

responsible

for

their

AI

system

and

its

(prompted)

output.

−  prohibits  its  Apertus  LLM  being  used  for  copyright  infringing  uses  (for  details

please  refer  to  chapter  A,  above).

4)  Contact  for  the  lodging  of  complaints
Affected  right  holders,  i.e.  right  holders  whose  copyrighted  material  has  been  used  for
training

purposes

of

the

Apertus

LLM

and

who

believe

their

copyright

has

been

infringed

during

this

training

process,

may

lodge

a

complaint

to:

llm-copyright-requests@swiss-ai.org
Complaints  must  be  sufficiently  precise  and  adequately  substantiated  regarding  the
non-compliance

of

SNAI

with

its

commitments

pursuant

to

this

chapter

B

and

provide

easily

accessible

information

about

it.

SNAI

will

act

within

reasonable

time

from

receiving

a

complaint

in

a

diligent

and

non-arbitrary

manner.

SNAI

reserves

the

right

not

to

respond

if

(i)

a

complaint

is

manifestly

unfounded

or

(ii)

has

already

been

addressed

to

an

identical

or

similar

complaint

by

the

same

rightsholder.

*******