engineering @mistralai, prev @tinybird

Paris
Joined May 2014
After 3 incredible years at @tinybird, it's time for my next adventure. I really enjoyed other people's tales of what companies are like from the inside, so here are my reflections on Tinybird: rbarbadillo.github.io/tinybi…
14
7
11
80
19,453
Raquel Barbadillo retweeted
if you want to work with us on training huge sparse MoEs (and get chonk cat plushies) join us! we are hiring a ton
really happy that we fully pre and post trained a 1T sized model training on 3800 grace blackwells, all in europe
95
27
4
1,071
21,233
Raquel Barbadillo retweeted
For their Mistral Large 4 release, @MistralAI ran a blind Surge human eval, with our expert software engineers evaluating five frontier models on coding quality. Le Chonk ranked #1 among open-weight models and #2 overall, behind only Opus 5. Unit tests ask: does it work? Human code review asks: would you merge it? Our evals measures the difference. Working code is the floor; professional taste and judgment are what make it shippable. Congrats to the Mistral team on le Chonk and their big release 🇫🇷
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
17
29
2
266
13,535
Raquel Barbadillo retweeted
Replying to @val_strch
la cadence va augmenter, on a fait beaucoup de progrès en infra
8
9
9
174
8,279
De todas las cosas que estamos publicando hoy, este hilo es el más interesante.
Today, we are launching a preview of our new model, Mistral Large 4 (ML4), aka le Chonk 🐈. ML4 is a 1T-parameter model with 49B active parameters, trained natively with multimodal capabilities. It is at the frontier of open models, and by far the strongest open-weight model from the US or Europe. The RL run behind this preview is still in flight and shows no sign of saturation -- we will release a final version before the end of the month along with the weights of the model. 🧵 1/n
1
1
23
903
Contratamos en ciencia para seguir construyendo más y mejores modelos. Contratamos en ingeniería para construir productos a su alrededor que sean útiles. Y en Mistral Compute, ya sabéis para qué :) Si estás interesado y nos conocemos, mándame un mensaje antes de aplicar: jobs.ashbyhq.com/mistral.ai
1
5
138
Raquel Barbadillo retweeted
This is a surprising great release. Le Chaton fat is real! Mistral Large 4 puts Europe back in contention on coding and cybersecurity. Didnt expect Mistral to compete with GLM5.3! “Le Chonk” is a natively multimodal model with 1T parameters and 49B active. The preview already delivers: - DeepSWE v1.1: 61.7%, roughly level with GLM-5.3 at 61%. Kimi K3 remains ahead at around 68% in Mistral’s comparison. - Artificial Analysis Cyber Index: 50, matching GLM-5.3-Flash and beating GLM-5.3’s 36. - CyberGym-E2E-AA: 82%, ahead of MiMo-V2.6-Pro’s 79%. Mistral says it trained the model from scratch in its own European datacenters. NVIDIA says only 4,000 Grace Blackwell Superchips powered the training! Which is impressive. Learning: you can compete with much less compute. Which is crazy. Congrats Mistral!
Today, we are launching a preview of our new model, Mistral Large 4 (ML4), aka le Chonk 🐈. ML4 is a 1T-parameter model with 49B active parameters, trained natively with multimodal capabilities. It is at the frontier of open models, and by far the strongest open-weight model from the US or Europe. The RL run behind this preview is still in flight and shows no sign of saturation -- we will release a final version before the end of the month along with the weights of the model. 🧵 1/n
58
47
9
742
42,790
Raquel Barbadillo retweeted
Genuinely super proud of the science team and everyone who contributed to this release. A lot is in store for products at Mistral and finally having a model this great will make everything come together.
3
2
44
1,155
Raquel Barbadillo retweeted
Some things to say about this model: 1. It's a good model in the actual usage sense. I've understood so much about Lacan and Merleau-Ponty in the last two days 2. I spent a non-trivial amount of time to make it good, and then to make it servable 3. We had a tattoo bet about its DeepSWE performance being over 60 (a capability I am responsible for) so I welcome tattoo parlour recommendations to send my colleagues to 4. I marvelled at how good some of my colleagues are at neural surgery under stress 5. An even chonkier boi is munching on trillions of tokens at this very moment mistral.ai/news/mistral-larg…
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
57
53
27
1,437
109,743
Raquel Barbadillo retweeted
Mistral Large 4 is now the #1 open-weight model on HLAB and #9 among open-weight models on the Vals Index.
16
37
10
399
28,863
let's gooooooooooooo
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
There's a new version of this post
1
96
Raquel Barbadillo retweeted
Replying to @arthurmensch
oh lawd he comin
2
3
286
Raquel Barbadillo retweeted
Replying to @arthurmensch
Go Arthur
3
20
626
Raquel Barbadillo retweeted
👀
2
10
57
1,259
Raquel Barbadillo retweeted
Excited
214
315
194
3,210
414,968
Raquel Barbadillo retweeted
the more compute used at write time, the less required at read time true in human communications & databases
11
24
5
394
22,551
Raquel Barbadillo retweeted
a lot of people are gonna wake up to a hard truth: years of teaching a personal agent how they think, work and make decisions are trapped inside a black box owned by someone else leaving means starting from scratch your agent will become the most valuable piece of software you own, make sure you actually own it
2
8
3
40
3,490
Raquel Barbadillo retweeted
Yo se que es IMPOSIBLE pero ya escribo desde la mas absoluta desesperación, estamos buscando piso con dos habitaciones para alquilar en Barcelona. Imploro que si sabeis algo me digais por favor, gracias!! 🙏🙏
10
19
1
33
11,818
esto es cierto de todo
📹 Juan Sanguino a Nicolás Coronado: “Hay barrios donde no parece realista ser una estrella de cine, pero si tu padre es José Coronado creces diciendo que ser actor es accesible”. @ZeroDramas_TVE analiza las ventajas de ser #nepobaby. #ZeroDramasLoles ⭕️rtve.es/play/videos/directo/…
3
1
1
1,837
Imagine you're a kid growing up in 1910. You happen to live next door to a doctor. Are you more likely to become a doctor by 1940 than your friend Carl, who lives 5 doors down? Our new paper suggests that yes, by about 30%!
61
Me da un poco de miedo el éxito de La Bola Negra por ver Costa Quebrada el verano que viene 🥺
1
3
430
(Más allá de eso loved it y realmente ojalá arrase en todos los lados)
134