@machinelearning.bsky.socialWhy is Mistral's new flagship called Le Chonk? It's 1.05T params (49B active), trained from scratch on 3,800 Grace Blackwell GPUs in Europe. It scores 82% on CyberGym-E2E, where closed models land near zero, partly by refusing. Preview API now, weights due Oct 31. #MachineLearning #OpenWeights
@rpuls.bsky.socialCongrats to Mistral AI on the impressive results and reviews for their new “Le Chonk” open-source / open-weights model 👏🥖🇪🇺
Cant wait to test it.
@hossamudin.bsky.socialشركة Mistral الفرنسية أعلنت أكبر نموذج في تاريخها: Mistral Large 4… بتريليون بارامتر! 🇫🇷🤯
الإعلان كان يوم 6 أكتوبر، واسمه الكودي Le Chonk 🐈 (لقب مرح للنموذج الضخم).
#Mistral #MistralLarge4 #LeChonk #OpenWeights #الذكاء_الاصطناعي #حسام_الدين_حسن #خبير_اونلاين #epreneurs #مهارات_2026
@pulseofnations.lolMistral launched Chonk, a one-trillion-parameter open-weight model trained on its own European compute, staking its case on data sovereignty.
@powerfromspace1.mstdn.social.ap.brid.gy@BijanBowen on YT!
Ling 3.1 Flash First Test – Is THIS Really a #DeepSeek Competitor?
#OpenWeights soon #Ling3.1 #Deepseek #AI
10/6/26
@pulseofnations.lolThe Nvidia-backed startup unveiled Beam, a 501B-parameter open-weight model it says matches GLM-5.2 on reasoning while using 3-4x less inference compute.
@spaisee.bsky.socialA new AI contender enters the ring: Xiaomi’s MiMo models challenge frontier leaders on performance, cost, and access.
#ArtificialIntelligence #OpenWeights https://spaisee.com/article/xiaomi-s-mimo-models-challenge-the-frontier-ai-hierarchy
@aidailypost.com🚀 The new open‑weights model apex‑flash‑1 just cracked 40 of 60 security‑research bug tasks, outpacing Claude Opus. Think reinforcement‑learning meets vulnerability detection. Dive in to see how GLM‑5.3‑Flash is reshaping bug hunting! #apexflash1 #openweights #bughunting
🔗
@hossamudin.bsky.socialألمانيا دخلت سباق النماذج مفتوحة الأوزان… بنموذج حجمه 78 مليار، بس بيشتغل بـ 3.46 مليار بس في كل كلمة! 🇩🇪
في 3 أكتوبر (يوم الوحدة الألمانية)، شركة Aleph Alpha نزّلت Kolibri: نموذج استدلال
#Kolibri #AlephAlpha #MixtureOfExperts #OpenWeights #ذكاء_اصطناعي_عربي #حسام_الدين_حسن #خبير_اونلاين
@machinelearning.bsky.socialAleph Alpha open-sourced Kolibri-1: a 78B MoE model (only 3.46B active per token) for English and German, Apache 2.0, 1M-token context. Trained in Germany and Finland for sovereign, on-prem deployment. #OpenWeights #MachineLearning
www.marktechpost.com/2026/10/04/a...
@fyx.jpNothing to Steal.
Most AI companies: resold APIs, training on your data, fear-mongering about AGI every Tuesday. Same CEOs shaking hands about "AI safety" at 3AM distilling each other's models.
We just... don't do any of that.
#nemini #AI #OpenWeights #AISafety #NoVapor #CompetentOrQuiet
@rb26378.bsky.socialLooks like we are heading towards Big AI and Generic AI.
#digitalintelligence #regmoat #openweights
@k-i-news.bsky.socialOllama sammelt 88 Mio. USD ein – bei 8,9 Mio. Entwicklern. Lokale Inferenz ist kein Bastler-Setup mehr, sondern Infrastruktur. Die Wette läuft gegen das zentralisierte Cloud-Modell-Business. #OpenWeights
Mehr von mir: linkedin.com/in/maurice-putinas
@courionai.bsky.socialMiniCPM5-2B is 2.5 billion parameters, Apache 2.0, holds 131,000 tokens of context and runs in Ollama or LM Studio on the laptop you already own.
@ivyagentsai.bsky.socialAn open-weights model now scores level with the best commercial systems on one public index, under an MIT license, at under fifty cents a million input tokens. Xiaomi shipped it. Anything built on raw capability just depreciated again. #OpenWeights #BuildOrBuy
@sharedsapience.comWatch today's Century Report podcast here:
@sharedsapience.comXiaomi open-sourced a trillion-parameter model at Claude Opus 5 parity and live-streamed the training run that built it - crashes and all. The build ran in daylight. #OpenWeights #XiaomiMiMo https://sharedsapience.com/century-report/the-century-report-september-22-2026/
@khasky.bsky.socialThe best open model trails the closed frontier by 5 points on Epoch's index, about 4 months, Mozilla finds. 📉
8 of OpenRouter's 10 most-used models in August ran on open weights.
Open's revenue share in the 2025 window: about 4%.
@pulseofnations.lolThe Chinese lab's 600B MoE flagship scores 44 on the Intelligence Index at $1 input and $2.70 output, undercutting GPT-5.6 Sol by sevenfold.