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-l…
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-l…
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 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.
71
58
1,717
147,133
Albert Jiang retweeted
@j_dekoninck Hi Jasper, I'm very sorry for this submission and the misleading impact it's had. There's no excuse for this oversight from me. We will rerun the model strictly following the rules.
2
1
26
1,934
Albert Jiang retweeted
We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next.
Using only 1,300 H200s, plus months of our experimental data, we mid-trained model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon.
This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials.
Read our blog posts below.
Using only 1,300 H200s, plus months of our experimental data, we mid-trained model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon.
This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials.
Read our blog posts below.
274
521
5,094
1,661,695
mathandai.org
This is actually a very good declaration. The diffusion of mathematics’s unreasonable effectiveness into reality requires understanding. But it’s absolutely mathematicians’ job to make this happen not AI companies cos there’s zero incentive for them beyond creating hype.
On the positive side, mathematics now has the chance to sieve out egoists and only welcome genuine appreciators.
This is actually a very good declaration. The diffusion of mathematics’s unreasonable effectiveness into reality requires understanding. But it’s absolutely mathematicians’ job to make this happen not AI companies cos there’s zero incentive for them beyond creating hype.
On the positive side, mathematics now has the chance to sieve out egoists and only welcome genuine appreciators.
2
1
14
1,997
matharena.ai/?comp=arxivlea…
A lot of grinding from Leanstral 1.5 with a tiny bit of K3 orchestration get to beat GPT-6 Astra on ArXivLean.
I think #tokens is counted erroneously here but yes a 6B model does yap. Great internship work by Matéo Pirio Rossignol mentored by @roman_soletskyi!
A lot of grinding from Leanstral 1.5 with a tiny bit of K3 orchestration get to beat GPT-6 Astra on ArXivLean.
I think #tokens is counted erroneously here but yes a 6B model does yap. Great internship work by Matéo Pirio Rossignol mentored by @roman_soletskyi!
3
8
54
5,231
That’s a nice view.
Wouldn’t it be nicer if it wasn’t from an oncology hospital; and I didn’t have to “donate” blood for an operation on my parent to go head; and the book I brought by chance wasn’t <<Never Let me Go>>
Wouldn’t it be nicer if it wasn’t from an oncology hospital; and I didn’t have to “donate” blood for an operation on my parent to go head; and the book I brought by chance wasn’t <<Never Let me Go>>
5
0
19
1,428
Chinese auntie in front me watching Memento for the entire 9 hours of the light and jumps back and forth to rewatch dialogues. Respect.
0
0
6
891
1. No one should do purely artisanal formalisation except for learning
2. It's insane if schools don't give STEM graduate students subscriptions and credits to coding agents
3. AI is good at forcing people to be honest. Deniers are delusional.
4. I think everything I wrote in March happened: albertqjiang.github.io/thoughts/agri-… Formalisation will be required in journals and for the sake of the community, invest in OS tools and models!
2. It's insane if schools don't give STEM graduate students subscriptions and credits to coding agents
3. AI is good at forcing people to be honest. Deniers are delusional.
4. I think everything I wrote in March happened: albertqjiang.github.io/thoughts/agri-… Formalisation will be required in journals and for the sake of the community, invest in OS tools and models!
Checking that a major mathematical proof is correct can take years. Formalization—converting the mathematical reasoning into a form computer proof assistants like Lean can verify—can help.
Last month, Claude completed the first formalized proof of Fermat’s Last Theorem, one of s of all time. This was a project experts thought would take many years. It is the largest Lean proof ever written.
Fermat’s Last Theorem was first proven in 1995 by Sir Andrew Wiles, more than 350 years after it was conjectured. Our proof, which totals over 13 million lines of code, provides machine verification. More importantly, it proves over 29,000 other theorems that the proof requires, across many areas of math which had never before been formalized.
We see this as a major step in the long process of firming up the core of mathematical knowledge, building on work from three centuries of mathematicians and hundreds of contributors to Lean and Mathlib. We are optimistic that AI-assisted verification of mathematical proofs will help reduce the burden of refereeing mathematics in an era where more proofs are being produced than ever before.
You can read about the process on our Science Blog: anthropic.com/research/forma…
And see the complete proof on GitHub: github.com/anthropics/fer…
4
4
52
6,644
Me to friends: Cambridge is not at all as quaint and idyllic as you think
Cambridge: So I'm just gonna put some horse drawn carriages in the park and use a phonebooth as the university library's book drop
Cambridge: So I'm just gonna put some horse drawn carriages in the park and use a phonebooth as the university library's book drop
1
0
9
947
One thing I used to (pre-GPT3.5) baptise as a symbol of agency was to start the proposed experiment before the call finishes. I first saw @jessemhan do this and was seriously impressed.
Nowadays I hardly leave any experiment hanging post-call. Agents grant agency.
Nowadays I hardly leave any experiment hanging post-call. Agents grant agency.
1
0
13
1,623
Great to see @WendaLi8 talk about Leanstral in the ICML tutorial: icml.cc/virtual/2026/7…
Starting 2:09:00 you see Leanstral present an elegant 6-line proof using grind compared to Claude's 40-50 line slop and Aristotle's 180-line MCTS trace (alledgedly) :P
Leanstral: github.com/Wenda302/ICML2…
Claude: github.com/Wenda302/ICML2…
Aristotle:
github.com/Wenda302/ICML2…
Genuinely gobsmacked by the Aristotle proof taking three vertical screenshots to capture and the last one exceeds twitter photo limit lol
Starting 2:09:00 you see Leanstral present an elegant 6-line proof using grind compared to Claude's 40-50 line slop and Aristotle's 180-line MCTS trace (alledgedly) :P
Leanstral: github.com/Wenda302/ICML2…
Claude: github.com/Wenda302/ICML2…
Aristotle:
github.com/Wenda302/ICML2…
Genuinely gobsmacked by the Aristotle proof taking three vertical screenshots to capture and the last one exceeds twitter photo limit lol
3
7
42
3,862
