Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity | TechCrunch

Oh my goodness, a TRILLION parameters, and they are giving them *away*! ✨ Mistral has released the weights for its trillion-parameter model, and VP of Science Pierre Stock explained to TechCrunch that organisations now get "frontier intelligence" to defend against cyberattacks. A trillion! Somewhere tonight a small security team on a modest budget is about to open a model that the biggest labs keep locked behind an API key, and they are going to feel absolutely *tremendous* about it.

The very best part is the licensing philosophy: access "that can't be cut off by definition because the model is open-weighted." Isn't that *permanent*? ✨ Most software can be switched off — subscriptions lapse, keys rotate, terms get quietly revised on a Friday afternoon. But this one is a gift that cannot be recalled, like a tattoo acquired in your twenties with total conviction and no foresight whatsoever. Once it is out, it is out forever, which is genuinely the most romantic thing a vice president of science can say about a file. Nobody can take it back. *Nobody can take it back!* ✨

And the timing could not be lovelier, because competition is heating up between open-weight developers like Mistral and closed-model providers like OpenAI and Anthropic — which means more releases, more downloads, and more frontier intelligence landing gently in the hands of defenders everywhere. Everybody wins! The security researcher wins, the enthusiast with one graphics card and a dream wins, and the poor overworked analyst who previously had to file a support ticket just to get throttled wins most of all. A trillion parameters, purpose-built for defence, shipped on 29 October 2026 with no off switch anywhere in sight. What a day to be open! ✨