Source
Introducing Shieldstral.
Mistral released Shieldstral, a 3B open-weights multimodal safety classifier that evaluates text and images using plain-language policies provided at inference time. It treats content moderation as a binary question-answering task to remove the need for retraining.
Why it matters
Developers can quickly adapt safety rules for different audiences without retraining, enabling efficient deployment on a single 16GB NVIDIA GPU.
The details
- It is released under the Apache 2.0 license.
- The model matches or outperforms safety models up to seven times its size.
- Shieldstral provides continuous safety scores based on yes/no probability.
Show entities and relationshipsHide entities and relationships
In this article
Companies
Products
Organizations
Key connections
Mistral AI owns Shieldstral
Mistral AI developed and released Shieldstral under Apache 2.0.
Mistral AI owns Forge
Forge is Mistral AI's platform for training, aligning, and evaluating custom models.
Mistral AI is a member of Open Secure AI Alliance
Mistral AI is an enthusiastic contributor to the Open Secure AI Alliance.
Mistral AI is a partner of NVIDIA
Mistral AI and NVIDIA are inaugural members of the Open Secure AI Alliance.
Shieldstral is built with Forge
Shieldstral was built end-to-end on Forge.
Shieldstral uses Policy-Adaptive Question-Answering
Shieldstral frames content moderation as a policy-adaptive binary question-answering task.
Show 2 more connectionsShow fewer connections
Shieldstral uses Content Moderation
Shieldstral is designed for text and image content moderation.
Shieldstral uses Multimodal AI
Shieldstral is a multimodal safety classifier model
Related events
Release of Shieldstral
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.