GPAI Ledger › GPAI Training Transparency tracker (AI Accountability Lab (AIAL)) › Capture 18 Aug 2026

GPAI Training Transparency tracker — capture 20260818T164910Z

Verify: sha256sum dd7c0257a731de5c56a6237243f64dd04c0f9ab811b0813f919a38e91f36589d.html.txt must equal the hash above (the filename IS the expected hash); ots verify dd7c0257a731de5c56a6237243f64dd04c0f9ab811b0813f919a38e91f36589d.html.ots -f dd7c0257a731de5c56a6237243f64dd04c0f9ab811b0813f919a38e91f36589d.html.txt (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).

GPAI Training Transparency

AIAL

☰

Home

Analysis

Methodology

Recommendations

Blog

Contact

GPAI Training Transparency

Quality Assessment AI Act Article 53(1)(d) Public Summaries

The
AI Act's Article 53(1)(d)
 requires General-Purpose AI (GPAI) model providers to publish
"a sufficiently detailed summary about the content used for training ... according to
a template
 provided by the
AI Office
"
. Enforcement by the AI Office has begun as of
2nd August 2026
.

To support this, we discovered
44 public summaries
, which we evaluate across two aspects:
Transparency
 and
Usefulness
 and assign a score using
our developed methodology
. Our findings expose challenges regarding accessibility, vagueness and incompleteness of information, and limitations this poses on rightsholders. We also highlight
missing summaries for 19 models
 based on our preliminary analysis. Please
get in touch
 if you find any summaries or would like to know more about our work.

See our peer-reviewed article: Dick A. H. Blankvoort, Harshvardhan J. Pandit, and Maximilian Gahntz (2026).
Quality Assessment of Public Summary of Training Content for GPAI models required by AI Act Article 53(1)(d)
. 9th ACM Conference on Fairness, Accountability, and Transparency (FAccT), Montreal, Canada. Zenodo.
DOI:10.1145/3805689.3806755

Media Coverage:

Euractiv
 "
AI labs at odds with EU over half-hearted data disclosures
 (7 Aug 2026)

Euractiv
 "
Researchers have trouble finding AI training data summaries
 (2 Mar 2026);

Tech Policy Press
 "
How Big AI Developers are Skirting a Mandate for Training Data Transparency
 (4 Mar 2026).

          The table below shows an
overview evaluated public summaries
 with
A+
 as the highest grade score and
F
 the lowest. You can also click on the columns to sort the table by name or by score. A separate list is provided further below with published summaries
 currently being evaluated
, and another list for models which require a
public summary is missing
 despite being necessary.
          You can click the model name to go to the detailed evaluation page which has more information, a link to the summary, and our evaluation notes. See
detailed overview
 with scores for each section of the public summary. The
list of summaries
 provides links to all found summaries.

Evaluated Public Summaries

Model

Provider

Transparency

Usefulness

Apertus

Swiss AI Initiative

                  Swiss AI Initiative

A

A+

FIBO

Bria AI

                  Bria AI

B+

A+

Bria 3.2

Bria AI

                  Bria AI

B+

A

SmolLM3-3B

HuggingFace

                  HuggingFace

B+

B+

Domyn Large

Domyn

                  Domyn

B+

B+

Bielik v3 11B Instruct

SpeakLeash

                  SpeakLeash

B+

C+

Adobe Firefly

Adobe

                  Adobe

C+

B+

Inkling

Thinking Machines

                  Thinking Machines

C+

C+

Inkling Small

Thinking Machines

                  Thinking Machines

C+

C+

FLUX.3

Black Forest Labs

                  Black Forest Labs

B

C

Nova 2 Lite

Amazon

                  Amazon

C+

C+

FastwebMIIA

Fastweb

                  Fastweb

C

C+

MAI Cyber 1 Flash

Microsoft

                  Microsoft

C+

C

Minimax M3

Minimax

                  Minimax

C+

D+

MAI Code 1 Flash

Microsoft

                  Microsoft

C+

C

MAI-Image-2.5

Microsoft

                  Microsoft

C+

D+

Ministral 3 14B

Mistral AI

                  Mistral AI

C

C

Ministral 3 3B

Mistral AI

                  Mistral AI

C

C

Ministral 3 8B

Mistral AI

                  Mistral AI

C

C

MAI-Image-2

Microsoft

                  Microsoft

C+

D+

GPT-5.6 Luna

OpenAI

                  OpenAI

C+

D+

Apertus v1.5

Swiss AI Initiative

                  Swiss AI Initiative

C

C+

GPT-5.5

OpenAI

                  OpenAI

C+

D+

Gemma 4

Google

                  Google

C

C

Gemini 3 Pro

Google

                  Google

C

C

Mistral Small 4

Mistral AI

                  Mistral AI

C

C

Mistral Large 3

Mistral AI

                  Mistral AI

C

C

Grok Voice Think Fast 2.0

xAI

                  xAI

C

C

Grok 4.5

xAI

                  xAI

C

C

C4AI Command A Plus

Cohere

                  Cohere

C

C

Claude Opus 5

Anthropic

                  Anthropic

C

D+

Claude Mythos Preview

Anthropic

                  Anthropic

C

D+

Claude Opus 4.7

Anthropic

                  Anthropic

C

D+

Claude Opus 4.8

Anthropic

                  Anthropic

C

D+

Claude Sonnet 5

Anthropic

                  Anthropic

C

D+

Claude Mythos 5 / Claude Fable 5

Anthropic

                  Anthropic

C

D+

Muse Image

Meta

                  Meta

C

D+

Muse Spark

Meta

                  Meta

C

D+

Phi-4

Microsoft

                  Microsoft

D

F

Summaries currently being evaluated

We are currently evaluating 5 public summaries. Their scores and our notes will be published shortly. In the meantime, you can access the public summary through the details page by clicking the model name.

Model

Provider

GPT Image 2

OpenAI

                  OpenAI

Muse Glimmer

Meta

                  Meta

GPT-5.4 Nano

OpenAI

                  OpenAI

GPT-5.2

OpenAI

                  OpenAI

PLLuM 2512 Base

Ministry of Digital Affairs of Poland

                  Ministry of Digital Affairs of Poland

Models with missing summaries

Based on our preliminary analysis, we have assessed the following models as requiring a mandatory public summary, but which we could not discover.

Model

Provider

FLUX.2 [max]

Black Forest Labs

                  Black Forest Labs

Granite 4.1

IBM

                  IBM

Veo 3.1

Google

                  Google

Palmyra X5

WRITER

                  WRITER

Claude Opus 4.6

Anthropic

                  Anthropic

Claude Sonnet 4.6

Anthropic

                  Anthropic

North Mini Code 1.0

Cohere

                  Cohere

FLUX.2

Black Forest Labs

                  Black Forest Labs

GPT-5.6 Sol

OpenAI

                  OpenAI

GPT-OSS

OpenAI

                  OpenAI

GPT-5.6 Terra

OpenAI

                  OpenAI

Apriel 1.5 15B Thinker

ServiceNow

                  ServiceNow

Claude Haiku 4.5

Anthropic

                  Anthropic

FLUX.2 Klein

Black Forest Labs

                  Black Forest Labs

Granite 4.0 H

IBM

                  IBM

Claude Opus 4.5

Anthropic

                  Anthropic

Sora 2

OpenAI

                  OpenAI

Claude Opus 4.1

Anthropic

                  Anthropic

Claude Sonnet 4.5

Anthropic

                  Anthropic

Why we started this project:

We contend that compliance cannot be
fait accompli
, and that the public summaries are a key factor in creating transparency and enabling rights enforcement. Towards this, our work also acts as a guide for providers who are yet to publish their summaries to consider how to do so with the highest possible quality and utility.

Compliance also invites practices that are intentionally or unintentionally deficient in achieving the goals. Our work serves as a useful tool for describing how and where and why certain practices are 'bad', e.g., where they use obfuscation, do not provide stated information. Using this, we can detect trends or patterns in whether the same issues occur in many summaries, and if so, how they can be collectively addressed through guidance, or enforced with priority.

The largest challenge in undertaking this work has been finding public summaries as there is no consistent format or practice for how they should be provided. For this, we provide
recommendations
.

The template for public summaries provided by the AI Office is intended to be revised with time to improve the state of documentation as well as to better guide the providers. We also provide
recommendations
 for these to improve the quality and accessibility of the public summaries.

This work has received funding from the Mozilla Foundation. The AIAL is supported by grants from following groups: the AI Collaborative, an Initiative of the Omidyar Group; Luminate; the Bestseller Foundation; the European Artificial Intelligence & Society Fund; and the John D. and Catherine T. MacArthur Foundation. Our Host Institute are: