TechPrufer
Get started
Mistral AI
Mistral AI logoMistral AI · Feature trail

Multimodal Safety Classification

5 posts · Sep 2024 – Aug 2026

What this trail covers

Shieldstral introduces a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size by framing content moderation as a policy-adaptive question-answering task. Unlike traditional guardrail models, it accepts plain-language policies at inference time, unifying text and image safety evaluation without retraining. Released under Apache 2.0, it delivers calibrated safety scores across diverse benchmarks while running efficiently on a single 16GB NVIDIA GPU. The Mistral Moderation API is the same API powering Le Chat's moderation service, offering raw text and conversational content classification endpoints. It is natively multilingual and trained on Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. The classifier addresses model-generated harms such as unqualified advice and PII.

All Mistral AI trails
  1. Introducing Shieldstral. | Mistral AIThis post details the introduction of Shieldstral, a 3B open-weights multimodal safety classifier. It frames content moderation as a policy-adaptive question-answering task, allowing plain-language policies at inference time without retraining. The model unifies text and image safety evaluation, providing calibrated safety scores. Key technical contributions include data unification from heterogeneous sources, training for discrimination rather than memorization using contrastive pairs, grounding safety in images with reranking, and merging complementary checkpoints (public data, generated data, base instruct model) via SLERP. The development was conducted on Forge, Mistral AI's platform for training, aligning, and evaluating custom models.
  2. [Deprecated] Pixtral Large | Mistral AIIntroduces Pixtral Large, a 124B multimodal model built on Mistral Large 2, offering frontier-level image understanding. It achieves state-of-the-art performance on MathVista, DocVQA, and VQAv2, and is competitive on MM-MT-Bench and the LMSys Vision Leaderboard. Pixtral Large features a 128K context window and demonstrates multilingual OCR and reasoning capabilities. Also announces an update to Mistral Large 24.11 with improvements in long context understanding, system prompt, and function calling.
  3. Mistral Moderation API | Mistral AIMistral AI releases a new content moderation API, which is the same API powering the moderation service in Le Chat. It offers two endpoints: one for raw text and one for conversational content, classifying the last message of a conversation within a conversational context. The model is natively multilingual and trained on 11 languages. It classifies text inputs into 9 categories and addresses model-generated harms like unqualified advice and PII. Performance is shared via AUC PR across policies on an internal testset.
  4. [Deprecated] Pixtral 12B | Mistral AIThis post announces Pixtral 12B, a natively multimodal model. It details its architecture, including a new 400M parameter vision encoder and a 12B parameter multimodal decoder based on Mistral Nemo. The model supports variable image sizes and aspect ratios, and can process multiple images within a 128k token context window. It demonstrates strong performance on multimodal reasoning benchmarks (MMMU), chart understanding, document question answering, and image-to-code generation, while also maintaining text-only benchmark performance. The post also notes that Pixtral 12B is deprecated and replaced by newer models.