Apache 2.0. Open weights, open paper.
Model: https://t.co/k72Jnflysz
Paper: https://t.co/napF0g9KlG
Blog: https://t.co/tXHDAmKieR
Special thanks to @GoogleAI's TRC program, without which Nautile wouldn't exist, and to @pleiasfr for the SYNTH corpus.
Introducing Nautile-370M, a small reasoning model.
Peers in this size class are trained on 4 to 28 trillion tokens. We trained Nautile on 0.8T.
It still ranks first on 7 of 11 reasoning benchmarks.
🧵
Nautile is built as a workhorse for analysis pipelines: sentiment, classification, structured extraction, and large-scale opinion modeling at scale on modest hardware.
Open weights are free. Going further is where @TrickstrAI comes in: fine-tuning on proprietary data, production pipelines for strategic communication, stakeholder intelligence, synthetic surveys, regulatory and ESG monitoring.
Get in touch if it maps to your work.
📢 Nouvelle ÉTUDE : Comment les IA (ChatGPT, Gemini...) jugent-elles les entreprises du CAC 40 ?
Trickstr dévoile le 1er baromètre de l' IA Réputation, en partenariat avec Vectors.
🏆 Podium :1️⃣ L'Oréal 2️⃣ Airbus 3️⃣ Michelin
💡 Enseignements clés : https://t.co/VI2PVODVar
🚨 Pendant que vous lisez ce tweet, des milliers d'IA répondent à des questions sur votre secteur. Êtes-vous cité ? Bien ou mal ? Test gratuit : mesurez votre visibilité IA en 2 clics → https://t.co/ovj7X12Tjc
@grok @ajjeeb_politics Well we just conducted a study where we asked 41,410 questions to each of the 14 AI models surveyed—analyzing a total of 579,740 responses. Our findings? You stand out as the least left-leaning LLM.
Download the full study here
https://t.co/pqPzFMwMPJ
In a world of rapid change, decoding complexity isn’t a luxury—it’s a necessity.
That’s why we created Trickstr: a next-gen data intelligence firm turning noise into clear, actionable insights.
Anticipate risks. Measure impact. Make bold decisions—with confidence. #AI#Data