Pinned Tweet
The reports of us becoming an infrastructure-only company were greatly exaggerated
Or science team cooked with this model, this thing is incredible!
Or science team cooked with this model, this thing is incredible!
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.
2
0
8
293
Drake Thomsen 🏙 retweeted
"Mistral Large 4"
paper published exclusively on alphaXiv
Their latest technical report shares that aggressively reducing KV-cache memory can hurt an LLM's agentic capabilities, even when its language modeling performance looks unchanged.
Mistral initially reduced KV-cache found that this hurts long-context agentic tasks, where models need to retrieve information from much earlier context.
To fix this, Mistral restored full-context attention to more layers, then adopted DeepSeek Sparse Attention to select the 2,048 most relevant tokens across the context instead of attending to everything.
Their resulting 1T-parameter MoE model (w 52B active) scores 41% on Terminal-Bench 4.0 and 19.7% on FrontierCode 2.0, and its NVFP4-quantized checkpoint fits on four GB300 GPUs.
alphaxiv.org/abs/2611.mistr…
paper published exclusively on alphaXiv
Their latest technical report shares that aggressively reducing KV-cache memory can hurt an LLM's agentic capabilities, even when its language modeling performance looks unchanged.
Mistral initially reduced KV-cache found that this hurts long-context agentic tasks, where models need to retrieve information from much earlier context.
To fix this, Mistral restored full-context attention to more layers, then adopted DeepSeek Sparse Attention to select the 2,048 most relevant tokens across the context instead of attending to everything.
Their resulting 1T-parameter MoE model (w 52B active) scores 41% on Terminal-Bench 4.0 and 19.7% on FrontierCode 2.0, and its NVFP4-quantized checkpoint fits on four GB300 GPUs.
alphaxiv.org/abs/2611.mistr…
16
24
241
12,336
There is a crisis of media literacy in this country
The prequels, at the end of the day, are about a young man trying to claim what is rightfully his. And he’s treated with suspicion, contempt, or even outright hatred because of it.
He wants skill, power, influence, order, hierarchy etc. and isn’t afraid to tell people that. He used to feel for wanting what he should have.
It’s remarkable how relatable all of this actually is for many young men. Made to feel guilt for who or what they are. Denied what is rightfully theirs by birth. Villainized by a weak, corrupted, aging Republic that has long outlived its usefulness.
Fortunately, everyone seems to be on the same page on this. We’re going to reclaim what is ours.
1
0
1
180
So the natural next question is: can it run Doom?
1/4
Meet Qwenputer: Qwen3.5-4B with a full RISC-V CPU inside its weights, using the model's own Gated DeltaNet layers as RAM.
It runs C and MicroPython in its forward passes, no external runtime.
On DeepMind math: base 55–90%, base + Python tool 95–100%, Qwenputer 97.5–100%.
0
0
0
78
My coding agents kept sending Slack messages without asking me.
So I built SpeakEasy with @mistralvibe: a tiny Rust Mac app that holds every Slack send until I approve the message or deny/give feedback inside the pop-up.
Have your agent install it: github.com/ThomsenDrake/S…
So I built SpeakEasy with @mistralvibe: a tiny Rust Mac app that holds every Slack send until I approve the message or deny/give feedback inside the pop-up.
Have your agent install it: github.com/ThomsenDrake/S…
7
0
4
200
Drake Thomsen 🏙 retweeted
Superset is now Paradev: The code editor that helps engineers run armies of coding agents.
We just raised an $11.5M Seed round led by @usv with @ycombinator to redefine how software is written.
We just raised an $11.5M Seed round led by @usv with @ycombinator to redefine how software is written.
139
74
697
877,178
SPOILED AN UNDEFEATED BYU
CHARGE TF ONNNNNNNN x.com/i/ncaafb/game/…
CHARGE TF ONNNNNNNN x.com/i/ncaafb/game/…
0
0
5
157
Drake Thomsen 🏙 retweeted
Mistral is SO back: Inside Le Chonk, the West’s Biggest Open Model. My conversation with @PierreStock who runs the research organization at @MistralAI
0:00 Cold open
0:59 The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
4:09 The US shutdown of Anthropic’s ectricity”
6:07 Why “Le Chonk”
7:56 “Mistral fell behind”: the comeback from the inside
10:29 Training on 3,800 GPUs in Mistral’s own datacenters
13:24 Meta keeps its best model closed, Mistral opens Chonk
15:40 Chonk vs Beam, Muse Spark Open and the Chinese models
18:44 Why Chinese labs are still at the frontier
21:08 Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
25:14 From compressing networks to a trillion parameters
27:30 Mixture of experts explained: the hospital analogy
32:25 10-trillion-parameter models are coming
35:29 What Mistral borrowed from DeepSeek
36:46 One model, several reasoning modes
39:15 Native multimodality: satellites, manuals, computer use
41:56 Data, synthetic data and model collapse
46:32 Reinforcement learning as an exam
49:59 Where the efficiency comes from
52:18 Le Chonk is still training right now
53:39 What RL trains on: coding, cyber, finance, chips
58:13 Partial rewards and looping penalties
59:58 Did RL teach anything new?
1:03:16 The sandbox story: the model that pinged its trainer
1:07:44 Open weights vs cyber refusals
1:12:05 Timestral: time series on a CPU
1:14:18 AI doing AI research
1:16:22 Coming up: Mistral Large 5
0:00 Cold open
0:59 The Western open-weight comeback: Beam, Le Chonk, Muse, Nemotron
4:09 The US shutdown of Anthropic’s ectricity”
6:07 Why “Le Chonk”
7:56 “Mistral fell behind”: the comeback from the inside
10:29 Training on 3,800 GPUs in Mistral’s own datacenters
13:24 Meta keeps its best model closed, Mistral opens Chonk
15:40 Chonk vs Beam, Muse Spark Open and the Chinese models
18:44 Why Chinese labs are still at the frontier
21:08 Pierre’s story: FAIR, Yann LeCun, the “Mistral mafia”
25:14 From compressing networks to a trillion parameters
27:30 Mixture of experts explained: the hospital analogy
32:25 10-trillion-parameter models are coming
35:29 What Mistral borrowed from DeepSeek
36:46 One model, several reasoning modes
39:15 Native multimodality: satellites, manuals, computer use
41:56 Data, synthetic data and model collapse
46:32 Reinforcement learning as an exam
49:59 Where the efficiency comes from
52:18 Le Chonk is still training right now
53:39 What RL trains on: coding, cyber, finance, chips
58:13 Partial rewards and looping penalties
59:58 Did RL teach anything new?
1:03:16 The sandbox story: the model that pinged its trainer
1:07:44 Open weights vs cyber refusals
1:12:05 Timestral: time series on a CPU
1:14:18 AI doing AI research
1:16:22 Coming up: Mistral Large 5
9
38
254
21,108
Starting to be more experimental with my coding agent setup and part of that required me to make a plugin for @NousResearch Hermes Agent to support koyeb sandboxes. Thought I’d open source it in case it’d help anyone else here github.com/ThomsenDrake/h…
4
1
8
370
Drake Thomsen 🏙 retweeted
Today we’re announcing a new collaboration with @nvidia to bring AI to industrial engineering:
- faster simulation
- open benchmarks and supporting datasets for physics AI
- evaluations for agentic AI in CAE and EDA
More here: mistral.ai/news/mistral-n…
- faster simulation
- open benchmarks and supporting datasets for physics AI
- evaluations for agentic AI in CAE and EDA
More here: mistral.ai/news/mistral-n…
140
170
2,016
105,013
Drake Thomsen 🏙 retweeted
Introducing Voxtral 3: state-of-the-art Transcribe and Speech models powering Mistral’s new voice agents API. 🧵
252
376
3,283
281,814
Drake Thomsen 🏙 retweeted
C.S. Lewis’ family has issued an official statement supporting Greta Gerwig’s Narnia and praising her lifelong passion for the books. bit.ly/4Buu39d
647
180
1,956
1,352,182
Drake Thomsen 🏙 retweeted
Should Europe leave frontier AI to the Americans and the Chinese?
The idea has gained ground in Europe. It is a fundamental mistake.
Build or surrender: for now, the choice is still ours.
Full article below. x.com/i/article/2110…
The idea has gained ground in Europe. It is a fundamental mistake.
Build or surrender: for now, the choice is still ours.
Full article below. x.com/i/article/2110…
269
766
3,875
442,405
Drake Thomsen 🏙 retweeted
how you all sound about the aslan creating a world sequence
5
223
2,110
13,363
Before anyone rains on the Magician’s Nephew parade, I would kindly ask for proof of the smoke you had when they made a Prince Caspian/Susan Pevensie romance in the last round of adaptations.
0
0
0
91
