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

Mistral has released the weights of a trillion-parameter AI model that the company is targeting at cybersecurity, TechCrunch reported on 29 October 2026. The report states that the release comes as competition increases between open-weight developers such as Mistral and closed-model providers such as OpenAI and Anthropic.

Pierre Stock, Mistral's vice president of science, told TechCrunch that organisations get access to frontier intelligence to defend against cyberattacks. The report quotes him describing that access as one "that can't be cut off by definition because the model is open-weighted." In the report's account, the open-weighted distribution is the reason the access cannot be withdrawn, and defence against cyberattacks is the purpose given for publishing the weights.

TechCrunch frames the release around a split between companies that publish model weights and companies that do not. Mistral is placed in the first group, with OpenAI and Anthropic named as closed-model providers in the second. The quoted rationale ties the cybersecurity purpose to the open-weighted form of the release rather than to a hosted arrangement. The report attributes that rationale to Stock, gives the model's size as one trillion parameters, and does not present an independent assessment of the cybersecurity claim.