Mistral releases open weights for Mistral Large 4 'Le Chonk'
Read original source →mistral.ai
The 1T-parameter MoE (49B active, 1M-token context) was trained in Europe and scores 82% on the Artificial Analysis Cyber Index. Mistral says three weeks of extra RL cut cost-per-task 63% and gave it a 1-point coding lead over GLM-5.3.
People · Pierre Stock, Sagar Vaze
How this story unfolded
5 days · 9 reports · 9 community posts · 18 of 20 shown
- Oct 29
Le Chonk: Built for Coders, Trusted by Defenders.mistral.ai
Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Racewired.com
Welcome Le Chonk: Mistral Large 4 on Hugging Facehuggingface.co
Mistral’s Large 4 open weights arrive with a 1-point coding lead over GLM-5.3: “There’s nothing you can’t do with this model that you can do with competitors.”thenewstack.io
Science Behind the Open Cyber Frontier with Mistral Large 4youtube.com
Mistral Large 4 updated weights and lower pricing on AI Gatewayvercel.com
Mistral releases weights for its trillion-parameter AI model, targeting cybersecuritytechcrunch.com
Mistral Large 4 'Le Chonk' arrives as the top open model outside China with improved coding performance and 63% lower cost-per-taskventurebeat.com
- Oct 30
- Nov 3
More stories today
Chatbots give error-laden answers on 2026 midterms, study finds
Forum AI found nearly 40% of election answers from ChatGPT, Claude and Gemini contained at least one factual error; Gemini gave a Pennsylvania mail-ballot deadline of November 3 instead of October 27. NewsGuard found chatbots cite "pink slime" partisan sites nearly 50% of the time on candidate questions.
Wired·1 hour ago

Ex-Legorans raise $2.4M pre-seed for real estate AI law firm Athena
Cem Ucan and Kieran John Keenan raised a $2.4M pre-seed led by a16z Speedrun for Athena, an AI-native law firm targeting commercial real estate. Proprietary agents review agreements, flag inconsistencies and draft redlines, with attorneys approving output; one purchase agreement turned around in under 48 hours.
Artificial Lawyer·1 hour ago

Cory Doctorow essay argues AI separates works from their makers
Pluralistic essay "Separable" argues AI produces paintings without painters, words without writers and mathematics without mathematicians. Doctorow grounds it in Edison's 1915-1925 touring "tone tests," where audiences mistook phonograph recordings for live performers.
Pluralistic·1 hour ago

CrowdStrike to buy Fiddler and Dreadnode for reported $495M
WSJ reports roughly $400M for Fiddler AI and about $95M for Dreadnode; CrowdStrike disclosed no terms and will pay mostly cash with some vesting stock. Fiddler's agent observability becomes Falcon Observe; Dreadnode runs 100+ AI attack strategies mapped to OWASP LLM Top 10 and MITRE ATLAS.
SecurityWeek·1 hour ago

Google.org funds five groups building AI tools for teachers
Google.org is providing over $9.5 million across five organizations building with and for educators. Recipients include CUNY's CITE initiative, Fab AI (with Google DeepMind and the Gates Foundation), Krause Center for Innovation, Teaching Lab's Coherence OS pilot, and 1EdTech Global.
Google Blog·1 hour ago

Google brings Guided Learning to Search and Classroom
Guided Learning, which coaches students through problems step by step instead of giving answers, is rolling out to Google Classroom globally and to Search for student accounts, starting with a pilot in select U.S. schools. Google is also updating school controls for Gemini, Gemini Notebook, Gemini in Classroom, and Search for K-12 accounts.
Google Blog·1 hour ago

DeepMind model wins 76.7% of simulated psychiatric intake cases
Kimmonismus·2 hours ago
Lovable CTO Fabian Hedin discusses AI-built software at Bloomberg Tech 2026
Lovable co-founder and CTO Fabian Hedin spoke with Bloomberg's Shirin Ghaffary at Bloomberg Tech 2026 in London about how AI is changing software creation and the role of engineers. The conversation also covered the challenge of building secure, maintainable applications with AI.
Bloomberg Technology·3 hours ago
