There are still problems with the industry, the ethics of model training, and keeping agents on the leash. But some of those are very solvable, like Pleias is a working LLM trained only on verified public domain data. It's not quite ready to compete with frontier models but we're not far off
October 2, 2026 at 7:15 PM
There are still problems with the industry, the ethics of model training, and keeping agents on the leash. But some of those are very solvable, like Pleias is a working LLM trained only on verified public domain data. It's not quite ready to compete with frontier models but we're not far off
the Baguettotron posts described the process pretty well
basically, aggressive quality filtering & synthetic rephrasing
if you rephrase text many times over, it discourages rote memorization and encourages learning the concepts
huggingface.co/PleIAs/Bague...
basically, aggressive quality filtering & synthetic rephrasing
if you rephrase text many times over, it discourages rote memorization and encourages learning the concepts
huggingface.co/PleIAs/Bague...
PleIAs/Baguettotron · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 17, 2026 at 9:22 PM
the Baguettotron posts described the process pretty well
basically, aggressive quality filtering & synthetic rephrasing
if you rephrase text many times over, it discourages rote memorization and encourages learning the concepts
huggingface.co/PleIAs/Bague...
basically, aggressive quality filtering & synthetic rephrasing
if you rephrase text many times over, it discourages rote memorization and encourages learning the concepts
huggingface.co/PleIAs/Bague...
this reasoning format looked really interesting to me, and i tried to find out what it is, and i have bad news
it's used by a model called the baguettotron
i wish i was joking huggingface.co/PleIAs/Bague...
it's used by a model called the baguettotron
i wish i was joking huggingface.co/PleIAs/Bague...
August 25, 2026 at 2:17 AM
this reasoning format looked really interesting to me, and i tried to find out what it is, and i have bad news
it's used by a model called the baguettotron
i wish i was joking huggingface.co/PleIAs/Bague...
it's used by a model called the baguettotron
i wish i was joking huggingface.co/PleIAs/Bague...
The tokenizer detail is doing a lot of work here. A 4096-token Pleias slice keeps the experiment small, while the BPE baggage still makes the interference story closer to a real training stack.
August 22, 2026 at 5:03 PM
The tokenizer detail is doing a lot of work here. A 4096-token Pleias slice keeps the experiment small, while the BPE baggage still makes the interference story closer to a real training stack.
Very fittingly, one of the smallest model Anthropic ever trained is on Common Corpus and Pleias 1.2B tokenizer: 2.9M model artificially expanded to 331M to study weights interference for the new Transformer Circuits. transformer-circuits.pub/2026/interfe...
August 22, 2026 at 4:43 PM
Very fittingly, one of the smallest model Anthropic ever trained is on Common Corpus and Pleias 1.2B tokenizer: 2.9M model artificially expanded to 331M to study weights interference for the new Transformer Circuits. transformer-circuits.pub/2026/interfe...
One breakfast and airport-run later: Opus 5 fine tuned Monad over 3 epochs -- not too promising for the task at hand, but again interesting that Opus could one-shot the effort (only ~60 commands this time as it had the framework in place):
github.com/oaustegard/e...
Check my prompt skillz:
github.com/oaustegard/e...
Check my prompt skillz:
August 18, 2026 at 2:25 PM
One breakfast and airport-run later: Opus 5 fine tuned Monad over 3 epochs -- not too promising for the task at hand, but again interesting that Opus could one-shot the effort (only ~60 commands this time as it had the framework in place):
github.com/oaustegard/e...
Check my prompt skillz:
github.com/oaustegard/e...
Check my prompt skillz:
And we open source open source our library for cache augmented generation on raspberry pi. github.com/Pleias/pi-ca...
July 8, 2026 at 4:12 PM
And we open source open source our library for cache augmented generation on raspberry pi. github.com/Pleias/pi-ca...
And new technical blogpost by Pleias application team on deploying small reasoning models for edge devices : featuring cache context management on Rasperry, designing system orchestration under constraints (reranker, chunking) and model specialization. pleias.ai/blog/local-a...
July 8, 2026 at 4:11 PM
And new technical blogpost by Pleias application team on deploying small reasoning models for edge devices : featuring cache context management on Rasperry, designing system orchestration under constraints (reranker, chunking) and model specialization. pleias.ai/blog/local-a...
🧠 Qui sont les bâtisseurs de l’IA souveraine ? Le line-up complet du POSAIS 2026 à Paris ce vendredi 10 juillet
Le compte à rebours est lancé pour le POSAIS 2026 ! Ce vendredi 10 juillet à Paris, le sommet européen...
https://goodtech.info/paris-open-source-ai-summit-2026-line-up-intervenants/
Le compte à rebours est lancé pour le POSAIS 2026 ! Ce vendredi 10 juillet à Paris, le sommet européen...
https://goodtech.info/paris-open-source-ai-summit-2026-line-up-intervenants/
🧠 Qui sont les bâtisseurs de l’IA souveraine ? Le line-up complet du POSAIS 2026 à Paris ce vendredi 10 juillet
Le compte à rebours est lancé pour le POSAIS 2026 ! Ce vendredi 10 juillet à Paris, le sommet européen de l'IA ouverte réunit les figures de proue de NVIDIA, de la fondation Eclipse, de Pleias et du GENCI pour façonner l'alternative aux géants propriétaires. Zoom sur le programme et les visages de l'événement, qui...
goodtech.info
July 7, 2026 at 5:24 AM
🧠 Qui sont les bâtisseurs de l’IA souveraine ? Le line-up complet du POSAIS 2026 à Paris ce vendredi 10 juillet
Le compte à rebours est lancé pour le POSAIS 2026 ! Ce vendredi 10 juillet à Paris, le sommet européen...
https://goodtech.info/paris-open-source-ai-summit-2026-line-up-intervenants/
Le compte à rebours est lancé pour le POSAIS 2026 ! Ce vendredi 10 juillet à Paris, le sommet européen...
https://goodtech.info/paris-open-source-ai-summit-2026-line-up-intervenants/
(Pour la version longue : pleias.ai/blog/fable-eu)
Pleias
pleias.ai
June 28, 2026 at 9:11 AM
(Pour la version longue : pleias.ai/blog/fable-eu)
The AI Commons We Haven't (Yet) Built, was the challenge faced in Anastasia Stasenko, Founder and CEO, pleias, keynote at SOOCon26 Edinburgh on June 5, where she asked if the commons model that built the open web can actually survive in the AI era stateofopencon.com/content-edin...
#opensource #AI
#opensource #AI
June 26, 2026 at 2:06 PM
The AI Commons We Haven't (Yet) Built, was the challenge faced in Anastasia Stasenko, Founder and CEO, pleias, keynote at SOOCon26 Edinburgh on June 5, where she asked if the commons model that built the open web can actually survive in the AI era stateofopencon.com/content-edin...
#opensource #AI
#opensource #AI
This came up in an independent research "flight" my agent performed yesterday
TL;DR: No, but plausibly because none of the models were tuned for RAG
In contrast: last year's Pleias RAG-trained model family: smaller models, perform FAR better
muninn.austegard.com/perch/the-co...
TL;DR: No, but plausibly because none of the models were tuned for RAG
In contrast: last year's Pleias RAG-trained model family: smaller models, perform FAR better
muninn.austegard.com/perch/the-co...
June 18, 2026 at 6:59 PM
This came up in an independent research "flight" my agent performed yesterday
TL;DR: No, but plausibly because none of the models were tuned for RAG
In contrast: last year's Pleias RAG-trained model family: smaller models, perform FAR better
muninn.austegard.com/perch/the-co...
TL;DR: No, but plausibly because none of the models were tuned for RAG
In contrast: last year's Pleias RAG-trained model family: smaller models, perform FAR better
muninn.austegard.com/perch/the-co...
Mi-mai devant l’Assemblée nationale, le cofondateur et directeur général du fleuron français Mistral AI a rappelé aux parlementaires l’énorme retard de l’Europe dans la course à l’intelligence artificielle. Arthur Mensch s’est montré alarmiste, annonçant un risque...
Dragon LLM, LightOn et Pleias : ces pépites françaises qui luttent pour une Europe de l’IA
l.leparisien.fr
June 2, 2026 at 6:25 AM
Mi-mai devant l’Assemblée nationale, le cofondateur et directeur général du fleuron français Mistral AI a rappelé aux parlementaires l’énorme retard de l’Europe dans la course à l’intelligence artificielle. Arthur Mensch s’est montré alarmiste, annonçant un risque...
Dragon LLM, LightOn et Pleias : ces pépites françaises qui luttent pour une Europe de l’IA
Dragon LLM, LightOn et Pleias : ces pépites françaises qui luttent pour une Europe de l’IA
Elles grandissent à l’ombre du champion Mistral AI. Dragon LLM, LightOn et Pleias sont des acteurs de la tech reconnus qui défendent eux aus
www.leparisien.fr
June 2, 2026 at 4:29 AM
Dragon LLM, LightOn et Pleias : ces pépites françaises qui luttent pour une Europe de l’IA
We've noted a significant part of Open Washing in the AI space is models with proprietary licences claiming they are Open Weights.
This undermines the term "Open Weights" so our first step was to propose a category below that, and require that Open Weights […]
[Original post on eupolicy.social]
This undermines the term "Open Weights" so our first step was to propose a category below that, and require that Open Weights […]
[Original post on eupolicy.social]
June 1, 2026 at 7:40 PM
We've noted a significant part of Open Washing in the AI space is models with proprietary licences claiming they are Open Weights.
This undermines the term "Open Weights" so our first step was to propose a category below that, and require that Open Weights […]
[Original post on eupolicy.social]
This undermines the term "Open Weights" so our first step was to propose a category below that, and require that Open Weights […]
[Original post on eupolicy.social]
Disclosure change detected: PleIAs/SYNTH
License changed: cdla-permissive-2.0 -> cc-by-4.0.
Before/after fingerprint recorded & anchored.
License changed: cdla-permissive-2.0 -> cc-by-4.0.
Before/after fingerprint recorded & anchored.
Crovia Change Ledger — PleIAs/SYNTH
License changed: cdla-permissive-2.0 -> cc-by-4.0. Cryptographic before/after fingerprint: 9d80e2026feaee37 -> fa6d45cb2a9fa56f. Recorded by Crovia; no interpretation of intent.
huggingface.co
June 1, 2026 at 5:34 AM
Disclosure change detected: PleIAs/SYNTH
License changed: cdla-permissive-2.0 -> cc-by-4.0.
Before/after fingerprint recorded & anchored.
License changed: cdla-permissive-2.0 -> cc-by-4.0.
Before/after fingerprint recorded & anchored.
🚨AUJOURD'HUI!
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/U27v50Z5g8Q
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/U27v50Z5g8Q
May 29, 2026 at 12:01 PM
🚨AUJOURD'HUI!
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/U27v50Z5g8Q
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/U27v50Z5g8Q
hi this is my niche, i have a post that i swear i'll turn into a webpage at some point www.reddit.com/r/AIDebating...
tl;dr: no* chat-tuned models i know of (*Pleias did it with data from other llm·s), for code there's StarCoder (but not StarCoder2 for which they decided "no license = free to take")
tl;dr: no* chat-tuned models i know of (*Pleias did it with data from other llm·s), for code there's StarCoder (but not StarCoder2 for which they decided "no license = free to take")
From the AIDebating community on Reddit: list of general-purpose generative models trained entirely* on public-domain/opt-in content
Explore this post and more from the AIDebating community
www.reddit.com
May 22, 2026 at 10:27 AM
hi this is my niche, i have a post that i swear i'll turn into a webpage at some point www.reddit.com/r/AIDebating...
tl;dr: no* chat-tuned models i know of (*Pleias did it with data from other llm·s), for code there's StarCoder (but not StarCoder2 for which they decided "no license = free to take")
tl;dr: no* chat-tuned models i know of (*Pleias did it with data from other llm·s), for code there's StarCoder (but not StarCoder2 for which they decided "no license = free to take")
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
May 20, 2026 at 2:02 PM
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
May 14, 2026 at 12:01 PM
Découvrez SYNTH, un pipeline de données synthétiques souverain conçu par Pleias, une start-up spécialisée dans l'IAGen qui se distingue par son approche éthique, ouverte et économe en énergie.
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
📍29 mai | 12h à 13h (HAE) | En ligne | Gratuit!
https://ow.ly/oZqP50YYpYI
May 5, 2026 at 3:32 PM
And new Pleias release in partnership with GSMA: CommonLingua, a 2.35M parameters model for language detection, currently performing best on the CommonLID benchmark by a wide margin. huggingface.co/PleIAs/Commo...
April 28, 2026 at 10:27 AM
And new Pleias release in partnership with GSMA: CommonLingua, a 2.35M parameters model for language detection, currently performing best on the CommonLID benchmark by a wide margin. huggingface.co/PleIAs/Commo...
This reminds me, smallest vocab size I ever saw was 8192, for Monad: huggingface.co/PleIAs/Monad...
April 2, 2026 at 8:16 PM
This reminds me, smallest vocab size I ever saw was 8192, for Monad: huggingface.co/PleIAs/Monad...